]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
78c8df12 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "classifier.h" | |
064af421 BP |
19 | #include <errno.h> |
20 | #include <netinet/in.h> | |
844dff32 | 21 | #include "byte-order.h" |
68d1c8c3 | 22 | #include "dynamic-string.h" |
064af421 BP |
23 | #include "flow.h" |
24 | #include "hash.h" | |
f2c21402 | 25 | #include "cmap.h" |
52054c15 | 26 | #include "list.h" |
07b37e8f | 27 | #include "odp-util.h" |
d8ae4d67 | 28 | #include "ofp-util.h" |
13751fd8 | 29 | #include "packets.h" |
52054c15 JR |
30 | #include "tag.h" |
31 | #include "util.h" | |
13751fd8 JR |
32 | #include "vlog.h" |
33 | ||
34 | VLOG_DEFINE_THIS_MODULE(classifier); | |
064af421 | 35 | |
69d6040e JR |
36 | struct trie_ctx; |
37 | ||
38 | /* Ports trie depends on both ports sharing the same ovs_be32. */ | |
39 | #define TP_PORTS_OFS32 (offsetof(struct flow, tp_src) / 4) | |
40 | BUILD_ASSERT_DECL(TP_PORTS_OFS32 == offsetof(struct flow, tp_dst) / 4); | |
cabd4c43 | 41 | |
cabd4c43 JR |
42 | /* A set of rules that all have the same fields wildcarded. */ |
43 | struct cls_subtable { | |
e65413ab | 44 | /* The fields are only used by writers and iterators. */ |
e48eccd1 | 45 | struct cmap_node cmap_node; /* Within struct classifier 'subtables_map'. */ |
e65413ab JR |
46 | |
47 | /* The fields are only used by writers. */ | |
48 | int n_rules OVS_GUARDED; /* Number of rules, including | |
49 | * duplicates. */ | |
50 | unsigned int max_priority OVS_GUARDED; /* Max priority of any rule in | |
51 | * the subtable. */ | |
52 | unsigned int max_count OVS_GUARDED; /* Count of max_priority rules. */ | |
53 | ||
54 | /* These fields are accessed by readers who care about wildcarding. */ | |
55 | tag_type tag; /* Tag generated from mask for partitioning (const). */ | |
56 | uint8_t n_indices; /* How many indices to use (const). */ | |
57 | uint8_t index_ofs[CLS_MAX_INDICES]; /* u32 segment boundaries (const). */ | |
58 | unsigned int trie_plen[CLS_MAX_TRIES]; /* Trie prefix length in 'mask' | |
59 | * (runtime configurable). */ | |
60 | int ports_mask_len; /* (const) */ | |
61 | struct cmap indices[CLS_MAX_INDICES]; /* Staged lookup indices. */ | |
f358a2cb | 62 | rcu_trie_ptr ports_trie; /* NULL if none. */ |
e65413ab JR |
63 | |
64 | /* These fields are accessed by all readers. */ | |
65 | struct cmap rules; /* Contains "struct cls_rule"s. */ | |
66 | struct minimask mask; /* Wildcards for fields (const). */ | |
3016f3e4 | 67 | /* 'mask' must be the last field. */ |
cabd4c43 JR |
68 | }; |
69 | ||
70 | /* Associates a metadata value (that is, a value of the OpenFlow 1.1+ metadata | |
71 | * field) with tags for the "cls_subtable"s that contain rules that match that | |
72 | * metadata value. */ | |
73 | struct cls_partition { | |
e48eccd1 | 74 | struct cmap_node cmap_node; /* In struct classifier's 'partitions' map. */ |
cabd4c43 JR |
75 | ovs_be64 metadata; /* metadata value for this partition. */ |
76 | tag_type tags; /* OR of each flow's cls_subtable tag. */ | |
e65413ab | 77 | struct tag_tracker tracker OVS_GUARDED; /* Tracks the bits in 'tags'. */ |
cabd4c43 JR |
78 | }; |
79 | ||
627fb667 JR |
80 | /* Internal representation of a rule in a "struct cls_subtable". */ |
81 | struct cls_match { | |
e65413ab JR |
82 | /* Accessed only by writers and iterators. */ |
83 | struct list list OVS_GUARDED; /* List of identical, lower-priority rules. */ | |
84 | ||
85 | /* Accessed only by writers. */ | |
86 | struct cls_partition *partition OVS_GUARDED; | |
87 | ||
88 | /* Accessed by readers interested in wildcarding. */ | |
89 | unsigned int priority; /* Larger numbers are higher priorities. */ | |
f2c21402 JR |
90 | struct cmap_node index_nodes[CLS_MAX_INDICES]; /* Within subtable's |
91 | * 'indices'. */ | |
e65413ab | 92 | /* Accessed by all readers. */ |
f2c21402 | 93 | struct cmap_node cmap_node; /* Within struct cls_subtable 'rules'. */ |
e65413ab | 94 | struct cls_rule *cls_rule; |
3016f3e4 JR |
95 | struct miniflow flow; /* Matching rule. Mask is in the subtable. */ |
96 | /* 'flow' must be the last field. */ | |
627fb667 | 97 | }; |
cabd4c43 | 98 | |
627fb667 JR |
99 | static struct cls_match * |
100 | cls_match_alloc(struct cls_rule *rule) | |
101 | { | |
3016f3e4 JR |
102 | int count = count_1bits(rule->match.flow.map); |
103 | ||
104 | struct cls_match *cls_match | |
105 | = xmalloc(sizeof *cls_match - sizeof cls_match->flow.inline_values | |
106 | + MINIFLOW_VALUES_SIZE(count)); | |
627fb667 JR |
107 | |
108 | cls_match->cls_rule = rule; | |
3016f3e4 | 109 | miniflow_clone_inline(&cls_match->flow, &rule->match.flow, count); |
627fb667 JR |
110 | cls_match->priority = rule->priority; |
111 | rule->cls_match = cls_match; | |
112 | ||
113 | return cls_match; | |
114 | } | |
cabd4c43 | 115 | |
e48eccd1 | 116 | static struct cls_subtable *find_subtable(const struct classifier *cls, |
d2064437 | 117 | const struct minimask *); |
e48eccd1 | 118 | static struct cls_subtable *insert_subtable(struct classifier *cls, |
e65413ab JR |
119 | const struct minimask *) |
120 | OVS_REQUIRES(cls->mutex); | |
e48eccd1 | 121 | static void destroy_subtable(struct classifier *cls, struct cls_subtable *) |
e65413ab | 122 | OVS_REQUIRES(cls->mutex); |
e48eccd1 | 123 | static struct cls_match *insert_rule(struct classifier *cls, |
e65413ab JR |
124 | struct cls_subtable *, struct cls_rule *) |
125 | OVS_REQUIRES(cls->mutex); | |
b5d97350 | 126 | |
627fb667 JR |
127 | static struct cls_match *find_match_wc(const struct cls_subtable *, |
128 | const struct flow *, struct trie_ctx *, | |
129 | unsigned int n_tries, | |
130 | struct flow_wildcards *); | |
131 | static struct cls_match *find_equal(struct cls_subtable *, | |
132 | const struct miniflow *, uint32_t hash); | |
b5d97350 | 133 | |
e65413ab JR |
134 | /* Iterates RULE over HEAD and all of the cls_rules on HEAD->list. |
135 | * Classifier's mutex must be held while iterating, as the list is | |
136 | * protoceted by it. */ | |
b5d97350 BP |
137 | #define FOR_EACH_RULE_IN_LIST(RULE, HEAD) \ |
138 | for ((RULE) = (HEAD); (RULE) != NULL; (RULE) = next_rule_in_list(RULE)) | |
139 | #define FOR_EACH_RULE_IN_LIST_SAFE(RULE, NEXT, HEAD) \ | |
140 | for ((RULE) = (HEAD); \ | |
141 | (RULE) != NULL && ((NEXT) = next_rule_in_list(RULE), true); \ | |
142 | (RULE) = (NEXT)) | |
143 | ||
627fb667 JR |
144 | static struct cls_match *next_rule_in_list__(struct cls_match *); |
145 | static struct cls_match *next_rule_in_list(struct cls_match *); | |
13751fd8 JR |
146 | |
147 | static unsigned int minimask_get_prefix_len(const struct minimask *, | |
148 | const struct mf_field *); | |
e48eccd1 | 149 | static void trie_init(struct classifier *cls, int trie_idx, |
e65413ab JR |
150 | const struct mf_field *) |
151 | OVS_REQUIRES(cls->mutex); | |
13751fd8 | 152 | static unsigned int trie_lookup(const struct cls_trie *, const struct flow *, |
c0bfb650 | 153 | union mf_value *plens); |
f358a2cb | 154 | static unsigned int trie_lookup_value(const rcu_trie_ptr *, |
c0bfb650 JR |
155 | const ovs_be32 value[], ovs_be32 plens[], |
156 | unsigned int value_bits); | |
f358a2cb | 157 | static void trie_destroy(rcu_trie_ptr *); |
13751fd8 | 158 | static void trie_insert(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 159 | static void trie_insert_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 160 | int mlen); |
13751fd8 | 161 | static void trie_remove(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 162 | static void trie_remove_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 163 | int mlen); |
13751fd8 | 164 | static void mask_set_prefix_bits(struct flow_wildcards *, uint8_t be32ofs, |
c30cfa6b | 165 | unsigned int n_bits); |
13751fd8 | 166 | static bool mask_prefix_bits_set(const struct flow_wildcards *, |
c30cfa6b | 167 | uint8_t be32ofs, unsigned int n_bits); |
3d91d909 JR |
168 | \f |
169 | /* flow/miniflow/minimask/minimatch utilities. | |
170 | * These are only used by the classifier, so place them here to allow | |
171 | * for better optimization. */ | |
172 | ||
173 | static inline uint64_t | |
174 | miniflow_get_map_in_range(const struct miniflow *miniflow, | |
175 | uint8_t start, uint8_t end, unsigned int *offset) | |
176 | { | |
177 | uint64_t map = miniflow->map; | |
178 | *offset = 0; | |
179 | ||
180 | if (start > 0) { | |
181 | uint64_t msk = (UINT64_C(1) << start) - 1; /* 'start' LSBs set */ | |
182 | *offset = count_1bits(map & msk); | |
183 | map &= ~msk; | |
184 | } | |
185 | if (end < FLOW_U32S) { | |
186 | uint64_t msk = (UINT64_C(1) << end) - 1; /* 'end' LSBs set */ | |
187 | map &= msk; | |
188 | } | |
189 | return map; | |
190 | } | |
191 | ||
192 | /* Returns a hash value for the bits of 'flow' where there are 1-bits in | |
193 | * 'mask', given 'basis'. | |
194 | * | |
195 | * The hash values returned by this function are the same as those returned by | |
196 | * miniflow_hash_in_minimask(), only the form of the arguments differ. */ | |
197 | static inline uint32_t | |
198 | flow_hash_in_minimask(const struct flow *flow, const struct minimask *mask, | |
199 | uint32_t basis) | |
200 | { | |
27bbe15d | 201 | const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks); |
3d91d909 | 202 | const uint32_t *flow_u32 = (const uint32_t *)flow; |
27bbe15d | 203 | const uint32_t *p = mask_values; |
3d91d909 JR |
204 | uint32_t hash; |
205 | uint64_t map; | |
206 | ||
207 | hash = basis; | |
208 | for (map = mask->masks.map; map; map = zero_rightmost_1bit(map)) { | |
33c6a1b9 | 209 | hash = hash_add(hash, flow_u32[raw_ctz(map)] & *p++); |
3d91d909 JR |
210 | } |
211 | ||
33c6a1b9 | 212 | return hash_finish(hash, (p - mask_values) * 4); |
3d91d909 JR |
213 | } |
214 | ||
215 | /* Returns a hash value for the bits of 'flow' where there are 1-bits in | |
216 | * 'mask', given 'basis'. | |
217 | * | |
218 | * The hash values returned by this function are the same as those returned by | |
219 | * flow_hash_in_minimask(), only the form of the arguments differ. */ | |
220 | static inline uint32_t | |
221 | miniflow_hash_in_minimask(const struct miniflow *flow, | |
222 | const struct minimask *mask, uint32_t basis) | |
223 | { | |
27bbe15d JR |
224 | const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks); |
225 | const uint32_t *p = mask_values; | |
3d91d909 JR |
226 | uint32_t hash = basis; |
227 | uint32_t flow_u32; | |
228 | ||
229 | MINIFLOW_FOR_EACH_IN_MAP(flow_u32, flow, mask->masks.map) { | |
33c6a1b9 | 230 | hash = hash_add(hash, flow_u32 & *p++); |
3d91d909 JR |
231 | } |
232 | ||
33c6a1b9 | 233 | return hash_finish(hash, (p - mask_values) * 4); |
3d91d909 JR |
234 | } |
235 | ||
236 | /* Returns a hash value for the bits of range [start, end) in 'flow', | |
237 | * where there are 1-bits in 'mask', given 'hash'. | |
238 | * | |
239 | * The hash values returned by this function are the same as those returned by | |
240 | * minimatch_hash_range(), only the form of the arguments differ. */ | |
241 | static inline uint32_t | |
242 | flow_hash_in_minimask_range(const struct flow *flow, | |
243 | const struct minimask *mask, | |
244 | uint8_t start, uint8_t end, uint32_t *basis) | |
245 | { | |
27bbe15d | 246 | const uint32_t *mask_values = miniflow_get_u32_values(&mask->masks); |
3d91d909 JR |
247 | const uint32_t *flow_u32 = (const uint32_t *)flow; |
248 | unsigned int offset; | |
249 | uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end, | |
250 | &offset); | |
27bbe15d | 251 | const uint32_t *p = mask_values + offset; |
3d91d909 JR |
252 | uint32_t hash = *basis; |
253 | ||
254 | for (; map; map = zero_rightmost_1bit(map)) { | |
33c6a1b9 | 255 | hash = hash_add(hash, flow_u32[raw_ctz(map)] & *p++); |
3d91d909 JR |
256 | } |
257 | ||
258 | *basis = hash; /* Allow continuation from the unfinished value. */ | |
33c6a1b9 | 259 | return hash_finish(hash, (p - mask_values) * 4); |
3d91d909 JR |
260 | } |
261 | ||
262 | /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask. */ | |
263 | static inline void | |
264 | flow_wildcards_fold_minimask(struct flow_wildcards *wc, | |
265 | const struct minimask *mask) | |
266 | { | |
267 | flow_union_with_miniflow(&wc->masks, &mask->masks); | |
268 | } | |
269 | ||
270 | /* Fold minimask 'mask''s wildcard mask into 'wc's wildcard mask | |
271 | * in range [start, end). */ | |
272 | static inline void | |
273 | flow_wildcards_fold_minimask_range(struct flow_wildcards *wc, | |
274 | const struct minimask *mask, | |
275 | uint8_t start, uint8_t end) | |
276 | { | |
277 | uint32_t *dst_u32 = (uint32_t *)&wc->masks; | |
278 | unsigned int offset; | |
279 | uint64_t map = miniflow_get_map_in_range(&mask->masks, start, end, | |
280 | &offset); | |
27bbe15d | 281 | const uint32_t *p = miniflow_get_u32_values(&mask->masks) + offset; |
3d91d909 JR |
282 | |
283 | for (; map; map = zero_rightmost_1bit(map)) { | |
284 | dst_u32[raw_ctz(map)] |= *p++; | |
285 | } | |
286 | } | |
287 | ||
288 | /* Returns a hash value for 'flow', given 'basis'. */ | |
289 | static inline uint32_t | |
290 | miniflow_hash(const struct miniflow *flow, uint32_t basis) | |
291 | { | |
27bbe15d JR |
292 | const uint32_t *values = miniflow_get_u32_values(flow); |
293 | const uint32_t *p = values; | |
3d91d909 JR |
294 | uint32_t hash = basis; |
295 | uint64_t hash_map = 0; | |
296 | uint64_t map; | |
297 | ||
298 | for (map = flow->map; map; map = zero_rightmost_1bit(map)) { | |
299 | if (*p) { | |
33c6a1b9 | 300 | hash = hash_add(hash, *p); |
3d91d909 JR |
301 | hash_map |= rightmost_1bit(map); |
302 | } | |
303 | p++; | |
304 | } | |
33c6a1b9 JR |
305 | hash = hash_add(hash, hash_map); |
306 | hash = hash_add(hash, hash_map >> 32); | |
3d91d909 | 307 | |
33c6a1b9 | 308 | return hash_finish(hash, p - values); |
3d91d909 JR |
309 | } |
310 | ||
311 | /* Returns a hash value for 'mask', given 'basis'. */ | |
312 | static inline uint32_t | |
313 | minimask_hash(const struct minimask *mask, uint32_t basis) | |
314 | { | |
315 | return miniflow_hash(&mask->masks, basis); | |
316 | } | |
317 | ||
318 | /* Returns a hash value for 'match', given 'basis'. */ | |
319 | static inline uint32_t | |
320 | minimatch_hash(const struct minimatch *match, uint32_t basis) | |
321 | { | |
322 | return miniflow_hash(&match->flow, minimask_hash(&match->mask, basis)); | |
323 | } | |
324 | ||
325 | /* Returns a hash value for the bits of range [start, end) in 'minimatch', | |
326 | * given 'basis'. | |
327 | * | |
328 | * The hash values returned by this function are the same as those returned by | |
329 | * flow_hash_in_minimask_range(), only the form of the arguments differ. */ | |
330 | static inline uint32_t | |
331 | minimatch_hash_range(const struct minimatch *match, uint8_t start, uint8_t end, | |
332 | uint32_t *basis) | |
333 | { | |
334 | unsigned int offset; | |
335 | const uint32_t *p, *q; | |
336 | uint32_t hash = *basis; | |
337 | int n, i; | |
338 | ||
339 | n = count_1bits(miniflow_get_map_in_range(&match->mask.masks, start, end, | |
340 | &offset)); | |
27bbe15d JR |
341 | q = miniflow_get_u32_values(&match->mask.masks) + offset; |
342 | p = miniflow_get_u32_values(&match->flow) + offset; | |
3d91d909 JR |
343 | |
344 | for (i = 0; i < n; i++) { | |
33c6a1b9 | 345 | hash = hash_add(hash, p[i] & q[i]); |
3d91d909 JR |
346 | } |
347 | *basis = hash; /* Allow continuation from the unfinished value. */ | |
33c6a1b9 | 348 | return hash_finish(hash, (offset + n) * 4); |
3d91d909 JR |
349 | } |
350 | ||
81a76618 BP |
351 | \f |
352 | /* cls_rule. */ | |
b5d97350 | 353 | |
81a76618 | 354 | /* Initializes 'rule' to match packets specified by 'match' at the given |
5cb7a798 BP |
355 | * 'priority'. 'match' must satisfy the invariant described in the comment at |
356 | * the definition of struct match. | |
66642cb4 | 357 | * |
48d28ac1 BP |
358 | * The caller must eventually destroy 'rule' with cls_rule_destroy(). |
359 | * | |
81a76618 BP |
360 | * (OpenFlow uses priorities between 0 and UINT16_MAX, inclusive, but |
361 | * internally Open vSwitch supports a wider range.) */ | |
47284b1f | 362 | void |
81a76618 BP |
363 | cls_rule_init(struct cls_rule *rule, |
364 | const struct match *match, unsigned int priority) | |
47284b1f | 365 | { |
5cb7a798 BP |
366 | minimatch_init(&rule->match, match); |
367 | rule->priority = priority; | |
627fb667 | 368 | rule->cls_match = NULL; |
5cb7a798 BP |
369 | } |
370 | ||
371 | /* Same as cls_rule_init() for initialization from a "struct minimatch". */ | |
372 | void | |
373 | cls_rule_init_from_minimatch(struct cls_rule *rule, | |
374 | const struct minimatch *match, | |
375 | unsigned int priority) | |
376 | { | |
377 | minimatch_clone(&rule->match, match); | |
81a76618 | 378 | rule->priority = priority; |
627fb667 | 379 | rule->cls_match = NULL; |
685a51a5 JP |
380 | } |
381 | ||
48d28ac1 BP |
382 | /* Initializes 'dst' as a copy of 'src'. |
383 | * | |
b2c1f00b | 384 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ |
48d28ac1 BP |
385 | void |
386 | cls_rule_clone(struct cls_rule *dst, const struct cls_rule *src) | |
387 | { | |
5cb7a798 BP |
388 | minimatch_clone(&dst->match, &src->match); |
389 | dst->priority = src->priority; | |
627fb667 | 390 | dst->cls_match = NULL; |
48d28ac1 BP |
391 | } |
392 | ||
b2c1f00b BP |
393 | /* Initializes 'dst' with the data in 'src', destroying 'src'. |
394 | * | |
395 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ | |
396 | void | |
397 | cls_rule_move(struct cls_rule *dst, struct cls_rule *src) | |
398 | { | |
399 | minimatch_move(&dst->match, &src->match); | |
400 | dst->priority = src->priority; | |
627fb667 | 401 | dst->cls_match = NULL; |
b2c1f00b BP |
402 | } |
403 | ||
48d28ac1 BP |
404 | /* Frees memory referenced by 'rule'. Doesn't free 'rule' itself (it's |
405 | * normally embedded into a larger structure). | |
406 | * | |
407 | * ('rule' must not currently be in a classifier.) */ | |
408 | void | |
5cb7a798 | 409 | cls_rule_destroy(struct cls_rule *rule) |
48d28ac1 | 410 | { |
627fb667 | 411 | ovs_assert(!rule->cls_match); |
5cb7a798 | 412 | minimatch_destroy(&rule->match); |
48d28ac1 BP |
413 | } |
414 | ||
81a76618 BP |
415 | /* Returns true if 'a' and 'b' match the same packets at the same priority, |
416 | * false if they differ in some way. */ | |
193eb874 BP |
417 | bool |
418 | cls_rule_equal(const struct cls_rule *a, const struct cls_rule *b) | |
419 | { | |
5cb7a798 | 420 | return a->priority == b->priority && minimatch_equal(&a->match, &b->match); |
193eb874 BP |
421 | } |
422 | ||
81a76618 | 423 | /* Returns a hash value for 'rule', folding in 'basis'. */ |
57452fdc BP |
424 | uint32_t |
425 | cls_rule_hash(const struct cls_rule *rule, uint32_t basis) | |
426 | { | |
5cb7a798 | 427 | return minimatch_hash(&rule->match, hash_int(rule->priority, basis)); |
73f33563 BP |
428 | } |
429 | ||
81a76618 | 430 | /* Appends a string describing 'rule' to 's'. */ |
07b37e8f BP |
431 | void |
432 | cls_rule_format(const struct cls_rule *rule, struct ds *s) | |
433 | { | |
5cb7a798 | 434 | minimatch_format(&rule->match, s, rule->priority); |
064af421 | 435 | } |
3ca1de08 BP |
436 | |
437 | /* Returns true if 'rule' matches every packet, false otherwise. */ | |
438 | bool | |
439 | cls_rule_is_catchall(const struct cls_rule *rule) | |
440 | { | |
5cb7a798 | 441 | return minimask_is_catchall(&rule->match.mask); |
3ca1de08 | 442 | } |
064af421 BP |
443 | \f |
444 | /* Initializes 'cls' as a classifier that initially contains no classification | |
445 | * rules. */ | |
446 | void | |
e48eccd1 JR |
447 | classifier_init(struct classifier *cls, const uint8_t *flow_segments) |
448 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 449 | { |
e65413ab | 450 | ovs_mutex_init(&cls->mutex); |
e65413ab | 451 | ovs_mutex_lock(&cls->mutex); |
064af421 | 452 | cls->n_rules = 0; |
f2c21402 | 453 | cmap_init(&cls->subtables_map); |
fe7cfa5c | 454 | pvector_init(&cls->subtables); |
f2c21402 | 455 | cmap_init(&cls->partitions); |
476f36e8 JR |
456 | cls->n_flow_segments = 0; |
457 | if (flow_segments) { | |
458 | while (cls->n_flow_segments < CLS_MAX_INDICES | |
459 | && *flow_segments < FLOW_U32S) { | |
460 | cls->flow_segments[cls->n_flow_segments++] = *flow_segments++; | |
461 | } | |
462 | } | |
13751fd8 | 463 | cls->n_tries = 0; |
e65413ab JR |
464 | for (int i = 0; i < CLS_MAX_TRIES; i++) { |
465 | trie_init(cls, i, NULL); | |
466 | } | |
467 | ovs_mutex_unlock(&cls->mutex); | |
064af421 BP |
468 | } |
469 | ||
470 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
afae68b1 JR |
471 | * caller's responsibility. |
472 | * May only be called after all the readers have been terminated. */ | |
064af421 | 473 | void |
e48eccd1 JR |
474 | classifier_destroy(struct classifier *cls) |
475 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 476 | { |
e48eccd1 | 477 | if (cls) { |
78c8df12 BP |
478 | struct cls_partition *partition; |
479 | struct cls_subtable *subtable; | |
13751fd8 JR |
480 | int i; |
481 | ||
e65413ab | 482 | ovs_mutex_lock(&cls->mutex); |
13751fd8 | 483 | for (i = 0; i < cls->n_tries; i++) { |
f358a2cb | 484 | trie_destroy(&cls->tries[i].root); |
13751fd8 | 485 | } |
064af421 | 486 | |
6bc3bb82 | 487 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
03868246 | 488 | destroy_subtable(cls, subtable); |
064af421 | 489 | } |
f2c21402 | 490 | cmap_destroy(&cls->subtables_map); |
c906cedf | 491 | |
6bc3bb82 | 492 | CMAP_FOR_EACH (partition, cmap_node, &cls->partitions) { |
f2c21402 | 493 | ovsrcu_postpone(free, partition); |
c906cedf | 494 | } |
f2c21402 | 495 | cmap_destroy(&cls->partitions); |
cabd4c43 | 496 | |
fe7cfa5c | 497 | pvector_destroy(&cls->subtables); |
e65413ab JR |
498 | ovs_mutex_unlock(&cls->mutex); |
499 | ovs_mutex_destroy(&cls->mutex); | |
064af421 BP |
500 | } |
501 | } | |
502 | ||
13751fd8 | 503 | /* Set the fields for which prefix lookup should be performed. */ |
f358a2cb | 504 | bool |
e48eccd1 | 505 | classifier_set_prefix_fields(struct classifier *cls, |
13751fd8 JR |
506 | const enum mf_field_id *trie_fields, |
507 | unsigned int n_fields) | |
e48eccd1 | 508 | OVS_EXCLUDED(cls->mutex) |
13751fd8 | 509 | { |
f358a2cb | 510 | const struct mf_field * new_fields[CLS_MAX_TRIES]; |
abadfcb0 | 511 | struct mf_bitmap fields = MF_BITMAP_INITIALIZER; |
f358a2cb JR |
512 | int i, n_tries = 0; |
513 | bool changed = false; | |
13751fd8 | 514 | |
e65413ab | 515 | ovs_mutex_lock(&cls->mutex); |
f358a2cb | 516 | for (i = 0; i < n_fields && n_tries < CLS_MAX_TRIES; i++) { |
13751fd8 JR |
517 | const struct mf_field *field = mf_from_id(trie_fields[i]); |
518 | if (field->flow_be32ofs < 0 || field->n_bits % 32) { | |
519 | /* Incompatible field. This is the only place where we | |
520 | * enforce these requirements, but the rest of the trie code | |
521 | * depends on the flow_be32ofs to be non-negative and the | |
522 | * field length to be a multiple of 32 bits. */ | |
523 | continue; | |
524 | } | |
525 | ||
abadfcb0 | 526 | if (bitmap_is_set(fields.bm, trie_fields[i])) { |
13751fd8 JR |
527 | /* Duplicate field, there is no need to build more than |
528 | * one index for any one field. */ | |
529 | continue; | |
530 | } | |
abadfcb0 | 531 | bitmap_set1(fields.bm, trie_fields[i]); |
13751fd8 | 532 | |
f358a2cb JR |
533 | new_fields[n_tries] = NULL; |
534 | if (n_tries >= cls->n_tries || field != cls->tries[n_tries].field) { | |
535 | new_fields[n_tries] = field; | |
536 | changed = true; | |
537 | } | |
538 | n_tries++; | |
539 | } | |
540 | ||
541 | if (changed || n_tries < cls->n_tries) { | |
542 | struct cls_subtable *subtable; | |
543 | ||
544 | /* Trie configuration needs to change. Disable trie lookups | |
545 | * for the tries that are changing and wait all the current readers | |
546 | * with the old configuration to be done. */ | |
547 | changed = false; | |
548 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
549 | for (i = 0; i < cls->n_tries; i++) { | |
550 | if ((i < n_tries && new_fields[i]) || i >= n_tries) { | |
551 | if (subtable->trie_plen[i]) { | |
552 | subtable->trie_plen[i] = 0; | |
553 | changed = true; | |
554 | } | |
555 | } | |
556 | } | |
557 | } | |
558 | /* Synchronize if any readers were using tries. The readers may | |
559 | * temporarily function without the trie lookup based optimizations. */ | |
560 | if (changed) { | |
561 | /* ovsrcu_synchronize() functions as a memory barrier, so it does | |
562 | * not matter that subtable->trie_plen is not atomic. */ | |
563 | ovsrcu_synchronize(); | |
13751fd8 | 564 | } |
13751fd8 | 565 | |
f358a2cb JR |
566 | /* Now set up the tries. */ |
567 | for (i = 0; i < n_tries; i++) { | |
568 | if (new_fields[i]) { | |
569 | trie_init(cls, i, new_fields[i]); | |
570 | } | |
571 | } | |
572 | /* Destroy the rest, if any. */ | |
573 | for (; i < cls->n_tries; i++) { | |
574 | trie_init(cls, i, NULL); | |
575 | } | |
576 | ||
577 | cls->n_tries = n_tries; | |
578 | ovs_mutex_unlock(&cls->mutex); | |
579 | return true; | |
13751fd8 | 580 | } |
f358a2cb | 581 | |
e65413ab | 582 | ovs_mutex_unlock(&cls->mutex); |
f358a2cb | 583 | return false; /* No change. */ |
13751fd8 JR |
584 | } |
585 | ||
586 | static void | |
e48eccd1 | 587 | trie_init(struct classifier *cls, int trie_idx, const struct mf_field *field) |
e65413ab | 588 | OVS_REQUIRES(cls->mutex) |
13751fd8 JR |
589 | { |
590 | struct cls_trie *trie = &cls->tries[trie_idx]; | |
591 | struct cls_subtable *subtable; | |
592 | ||
593 | if (trie_idx < cls->n_tries) { | |
f358a2cb JR |
594 | trie_destroy(&trie->root); |
595 | } else { | |
596 | ovsrcu_set_hidden(&trie->root, NULL); | |
13751fd8 | 597 | } |
13751fd8 JR |
598 | trie->field = field; |
599 | ||
f358a2cb | 600 | /* Add existing rules to the new trie. */ |
f2c21402 | 601 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
13751fd8 JR |
602 | unsigned int plen; |
603 | ||
604 | plen = field ? minimask_get_prefix_len(&subtable->mask, field) : 0; | |
13751fd8 | 605 | if (plen) { |
627fb667 | 606 | struct cls_match *head; |
13751fd8 | 607 | |
f2c21402 | 608 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 609 | struct cls_match *match; |
13751fd8 | 610 | |
627fb667 JR |
611 | FOR_EACH_RULE_IN_LIST (match, head) { |
612 | trie_insert(trie, match->cls_rule, plen); | |
13751fd8 JR |
613 | } |
614 | } | |
615 | } | |
f358a2cb JR |
616 | /* Initialize subtable's prefix length on this field. This will |
617 | * allow readers to use the trie. */ | |
618 | atomic_thread_fence(memory_order_release); | |
619 | subtable->trie_plen[trie_idx] = plen; | |
13751fd8 JR |
620 | } |
621 | } | |
622 | ||
5f0476ce JR |
623 | /* Returns true if 'cls' contains no classification rules, false otherwise. |
624 | * Checking the cmap requires no locking. */ | |
064af421 BP |
625 | bool |
626 | classifier_is_empty(const struct classifier *cls) | |
627 | { | |
e48eccd1 | 628 | return cmap_is_empty(&cls->subtables_map); |
064af421 BP |
629 | } |
630 | ||
dbda2960 | 631 | /* Returns the number of rules in 'cls'. */ |
064af421 BP |
632 | int |
633 | classifier_count(const struct classifier *cls) | |
afae68b1 | 634 | OVS_NO_THREAD_SAFETY_ANALYSIS |
064af421 | 635 | { |
afae68b1 JR |
636 | /* n_rules is an int, so in the presence of concurrent writers this will |
637 | * return either the old or a new value. */ | |
e48eccd1 | 638 | return cls->n_rules; |
064af421 BP |
639 | } |
640 | ||
c906cedf BP |
641 | static uint32_t |
642 | hash_metadata(ovs_be64 metadata_) | |
643 | { | |
644 | uint64_t metadata = (OVS_FORCE uint64_t) metadata_; | |
965607c8 | 645 | return hash_uint64(metadata); |
c906cedf BP |
646 | } |
647 | ||
648 | static struct cls_partition * | |
e48eccd1 | 649 | find_partition(const struct classifier *cls, ovs_be64 metadata, uint32_t hash) |
c906cedf BP |
650 | { |
651 | struct cls_partition *partition; | |
652 | ||
f2c21402 | 653 | CMAP_FOR_EACH_WITH_HASH (partition, cmap_node, hash, &cls->partitions) { |
c906cedf BP |
654 | if (partition->metadata == metadata) { |
655 | return partition; | |
656 | } | |
657 | } | |
658 | ||
659 | return NULL; | |
660 | } | |
661 | ||
662 | static struct cls_partition * | |
e48eccd1 | 663 | create_partition(struct classifier *cls, struct cls_subtable *subtable, |
c906cedf | 664 | ovs_be64 metadata) |
e65413ab | 665 | OVS_REQUIRES(cls->mutex) |
c906cedf BP |
666 | { |
667 | uint32_t hash = hash_metadata(metadata); | |
668 | struct cls_partition *partition = find_partition(cls, metadata, hash); | |
669 | if (!partition) { | |
670 | partition = xmalloc(sizeof *partition); | |
671 | partition->metadata = metadata; | |
672 | partition->tags = 0; | |
183126a1 | 673 | tag_tracker_init(&partition->tracker); |
f2c21402 | 674 | cmap_insert(&cls->partitions, &partition->cmap_node, hash); |
c906cedf | 675 | } |
03868246 | 676 | tag_tracker_add(&partition->tracker, &partition->tags, subtable->tag); |
c906cedf BP |
677 | return partition; |
678 | } | |
679 | ||
69d6040e JR |
680 | static inline ovs_be32 minimatch_get_ports(const struct minimatch *match) |
681 | { | |
682 | /* Could optimize to use the same map if needed for fast path. */ | |
683 | return MINIFLOW_GET_BE32(&match->flow, tp_src) | |
684 | & MINIFLOW_GET_BE32(&match->mask.masks, tp_src); | |
685 | } | |
686 | ||
b5d97350 BP |
687 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
688 | * must not modify or free it. | |
064af421 BP |
689 | * |
690 | * If 'cls' already contains an identical rule (including wildcards, values of | |
691 | * fixed fields, and priority), replaces the old rule by 'rule' and returns the | |
692 | * rule that was replaced. The caller takes ownership of the returned rule and | |
48d28ac1 BP |
693 | * is thus responsible for destroying it with cls_rule_destroy(), freeing the |
694 | * memory block in which it resides, etc., as necessary. | |
064af421 BP |
695 | * |
696 | * Returns NULL if 'cls' does not contain a rule with an identical key, after | |
697 | * inserting the new rule. In this case, no rules are displaced by the new | |
698 | * rule, even rules that cannot have any effect because the new rule matches a | |
699 | * superset of their flows and has higher priority. */ | |
700 | struct cls_rule * | |
e48eccd1 JR |
701 | classifier_replace(struct classifier *cls, struct cls_rule *rule) |
702 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 703 | { |
627fb667 | 704 | struct cls_match *old_rule; |
03868246 | 705 | struct cls_subtable *subtable; |
e65413ab | 706 | struct cls_rule *old_cls_rule = NULL; |
b5d97350 | 707 | |
e65413ab | 708 | ovs_mutex_lock(&cls->mutex); |
03868246 JR |
709 | subtable = find_subtable(cls, &rule->match.mask); |
710 | if (!subtable) { | |
711 | subtable = insert_subtable(cls, &rule->match.mask); | |
b5d97350 BP |
712 | } |
713 | ||
03868246 | 714 | old_rule = insert_rule(cls, subtable, rule); |
b5d97350 | 715 | if (!old_rule) { |
e65413ab | 716 | old_cls_rule = NULL; |
13751fd8 | 717 | |
627fb667 | 718 | rule->cls_match->partition = NULL; |
c906cedf BP |
719 | if (minimask_get_metadata_mask(&rule->match.mask) == OVS_BE64_MAX) { |
720 | ovs_be64 metadata = miniflow_get_metadata(&rule->match.flow); | |
627fb667 JR |
721 | rule->cls_match->partition = create_partition(cls, subtable, |
722 | metadata); | |
c906cedf BP |
723 | } |
724 | ||
064af421 | 725 | cls->n_rules++; |
13751fd8 | 726 | |
e65413ab | 727 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
728 | if (subtable->trie_plen[i]) { |
729 | trie_insert(&cls->tries[i], rule, subtable->trie_plen[i]); | |
730 | } | |
731 | } | |
69d6040e JR |
732 | |
733 | /* Ports trie. */ | |
734 | if (subtable->ports_mask_len) { | |
735 | /* We mask the value to be inserted to always have the wildcarded | |
736 | * bits in known (zero) state, so we can include them in comparison | |
737 | * and they will always match (== their original value does not | |
738 | * matter). */ | |
739 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
740 | ||
741 | trie_insert_prefix(&subtable->ports_trie, &masked_ports, | |
742 | subtable->ports_mask_len); | |
743 | } | |
c906cedf | 744 | } else { |
e65413ab | 745 | old_cls_rule = old_rule->cls_rule; |
627fb667 JR |
746 | rule->cls_match->partition = old_rule->partition; |
747 | old_cls_rule->cls_match = NULL; | |
f2c21402 JR |
748 | |
749 | /* 'old_rule' contains a cmap_node, which may not be freed | |
750 | * immediately. */ | |
751 | ovsrcu_postpone(free, old_rule); | |
064af421 | 752 | } |
e65413ab JR |
753 | ovs_mutex_unlock(&cls->mutex); |
754 | return old_cls_rule; | |
064af421 BP |
755 | } |
756 | ||
08944c1d BP |
757 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
758 | * must not modify or free it. | |
759 | * | |
760 | * 'cls' must not contain an identical rule (including wildcards, values of | |
761 | * fixed fields, and priority). Use classifier_find_rule_exactly() to find | |
762 | * such a rule. */ | |
763 | void | |
764 | classifier_insert(struct classifier *cls, struct cls_rule *rule) | |
765 | { | |
766 | struct cls_rule *displaced_rule = classifier_replace(cls, rule); | |
cb22974d | 767 | ovs_assert(!displaced_rule); |
08944c1d BP |
768 | } |
769 | ||
48d28ac1 BP |
770 | /* Removes 'rule' from 'cls'. It is the caller's responsibility to destroy |
771 | * 'rule' with cls_rule_destroy(), freeing the memory block in which 'rule' | |
747f140a JR |
772 | * resides, etc., as necessary. |
773 | * | |
774 | * Does nothing if 'rule' has been already removed, or was never inserted. | |
775 | * | |
776 | * Returns the removed rule, or NULL, if it was already removed. | |
777 | */ | |
778 | struct cls_rule * | |
e48eccd1 JR |
779 | classifier_remove(struct classifier *cls, struct cls_rule *rule) |
780 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 781 | { |
c906cedf | 782 | struct cls_partition *partition; |
747f140a | 783 | struct cls_match *cls_match; |
627fb667 | 784 | struct cls_match *head; |
03868246 | 785 | struct cls_subtable *subtable; |
476f36e8 | 786 | int i; |
f2c21402 JR |
787 | uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES]; |
788 | uint8_t prev_be32ofs = 0; | |
064af421 | 789 | |
e65413ab | 790 | ovs_mutex_lock(&cls->mutex); |
747f140a JR |
791 | cls_match = rule->cls_match; |
792 | if (!cls_match) { | |
793 | rule = NULL; | |
794 | goto unlock; /* Already removed. */ | |
795 | } | |
796 | ||
03868246 | 797 | subtable = find_subtable(cls, &rule->match.mask); |
627fb667 JR |
798 | ovs_assert(subtable); |
799 | ||
69d6040e JR |
800 | if (subtable->ports_mask_len) { |
801 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
802 | ||
803 | trie_remove_prefix(&subtable->ports_trie, | |
804 | &masked_ports, subtable->ports_mask_len); | |
805 | } | |
13751fd8 JR |
806 | for (i = 0; i < cls->n_tries; i++) { |
807 | if (subtable->trie_plen[i]) { | |
808 | trie_remove(&cls->tries[i], rule, subtable->trie_plen[i]); | |
809 | } | |
810 | } | |
811 | ||
476f36e8 JR |
812 | /* Remove rule node from indices. */ |
813 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 JR |
814 | ihash[i] = minimatch_hash_range(&rule->match, prev_be32ofs, |
815 | subtable->index_ofs[i], &basis); | |
816 | cmap_remove(&subtable->indices[i], &cls_match->index_nodes[i], | |
817 | ihash[i]); | |
818 | prev_be32ofs = subtable->index_ofs[i]; | |
476f36e8 | 819 | } |
f2c21402 | 820 | hash = minimatch_hash_range(&rule->match, prev_be32ofs, FLOW_U32S, &basis); |
476f36e8 | 821 | |
f2c21402 | 822 | head = find_equal(subtable, &rule->match.flow, hash); |
627fb667 JR |
823 | if (head != cls_match) { |
824 | list_remove(&cls_match->list); | |
825 | } else if (list_is_empty(&cls_match->list)) { | |
f2c21402 | 826 | cmap_remove(&subtable->rules, &cls_match->cmap_node, hash); |
b5d97350 | 827 | } else { |
627fb667 JR |
828 | struct cls_match *next = CONTAINER_OF(cls_match->list.next, |
829 | struct cls_match, list); | |
064af421 | 830 | |
627fb667 | 831 | list_remove(&cls_match->list); |
f2c21402 JR |
832 | cmap_replace(&subtable->rules, &cls_match->cmap_node, |
833 | &next->cmap_node, hash); | |
b5d97350 | 834 | } |
064af421 | 835 | |
627fb667 | 836 | partition = cls_match->partition; |
183126a1 BP |
837 | if (partition) { |
838 | tag_tracker_subtract(&partition->tracker, &partition->tags, | |
03868246 | 839 | subtable->tag); |
183126a1 | 840 | if (!partition->tags) { |
f2c21402 JR |
841 | cmap_remove(&cls->partitions, &partition->cmap_node, |
842 | hash_metadata(partition->metadata)); | |
843 | ovsrcu_postpone(free, partition); | |
183126a1 | 844 | } |
c906cedf BP |
845 | } |
846 | ||
03868246 JR |
847 | if (--subtable->n_rules == 0) { |
848 | destroy_subtable(cls, subtable); | |
fe7cfa5c JR |
849 | } else if (subtable->max_priority == cls_match->priority |
850 | && --subtable->max_count == 0) { | |
851 | /* Find the new 'max_priority' and 'max_count'. */ | |
852 | struct cls_match *head; | |
853 | unsigned int max_priority = 0; | |
854 | ||
f2c21402 | 855 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
fe7cfa5c JR |
856 | if (head->priority > max_priority) { |
857 | max_priority = head->priority; | |
858 | subtable->max_count = 1; | |
859 | } else if (head->priority == max_priority) { | |
860 | ++subtable->max_count; | |
861 | } | |
862 | } | |
863 | subtable->max_priority = max_priority; | |
864 | pvector_change_priority(&cls->subtables, subtable, max_priority); | |
4d935a6b | 865 | } |
13751fd8 | 866 | |
b5d97350 | 867 | cls->n_rules--; |
627fb667 | 868 | |
f2c21402 | 869 | ovsrcu_postpone(free, cls_match); |
747f140a JR |
870 | rule->cls_match = NULL; |
871 | unlock: | |
e65413ab | 872 | ovs_mutex_unlock(&cls->mutex); |
747f140a JR |
873 | |
874 | return rule; | |
064af421 BP |
875 | } |
876 | ||
13751fd8 | 877 | /* Prefix tree context. Valid when 'lookup_done' is true. Can skip all |
c0bfb650 JR |
878 | * subtables which have a prefix match on the trie field, but whose prefix |
879 | * length is not indicated in 'match_plens'. For example, a subtable that | |
880 | * has a 8-bit trie field prefix match can be skipped if | |
881 | * !be_get_bit_at(&match_plens, 8 - 1). If skipped, 'maskbits' prefix bits | |
882 | * must be unwildcarded to make datapath flow only match packets it should. */ | |
13751fd8 JR |
883 | struct trie_ctx { |
884 | const struct cls_trie *trie; | |
885 | bool lookup_done; /* Status of the lookup. */ | |
886 | uint8_t be32ofs; /* U32 offset of the field in question. */ | |
13751fd8 | 887 | unsigned int maskbits; /* Prefix length needed to avoid false matches. */ |
c0bfb650 JR |
888 | union mf_value match_plens; /* Bitmask of prefix lengths with possible |
889 | * matches. */ | |
13751fd8 JR |
890 | }; |
891 | ||
892 | static void | |
893 | trie_ctx_init(struct trie_ctx *ctx, const struct cls_trie *trie) | |
894 | { | |
895 | ctx->trie = trie; | |
896 | ctx->be32ofs = trie->field->flow_be32ofs; | |
897 | ctx->lookup_done = false; | |
898 | } | |
899 | ||
48c3de13 BP |
900 | /* Finds and returns the highest-priority rule in 'cls' that matches 'flow'. |
901 | * Returns a null pointer if no rules in 'cls' match 'flow'. If multiple rules | |
74f74083 EJ |
902 | * of equal priority match 'flow', returns one arbitrarily. |
903 | * | |
904 | * If a rule is found and 'wc' is non-null, bitwise-OR's 'wc' with the | |
905 | * set of bits that were significant in the lookup. At some point | |
906 | * earlier, 'wc' should have been initialized (e.g., by | |
907 | * flow_wildcards_init_catchall()). */ | |
48c3de13 | 908 | struct cls_rule * |
e48eccd1 | 909 | classifier_lookup(const struct classifier *cls, const struct flow *flow, |
74f74083 | 910 | struct flow_wildcards *wc) |
48c3de13 | 911 | { |
c906cedf | 912 | const struct cls_partition *partition; |
c906cedf | 913 | tag_type tags; |
ec988646 | 914 | int64_t best_priority = -1; |
fe7cfa5c JR |
915 | const struct cls_match *best; |
916 | struct trie_ctx trie_ctx[CLS_MAX_TRIES]; | |
917 | struct cls_subtable *subtable; | |
c906cedf | 918 | |
f358a2cb JR |
919 | /* Synchronize for cls->n_tries and subtable->trie_plen. They can change |
920 | * when table configuration changes, which happens typically only on | |
921 | * startup. */ | |
922 | atomic_thread_fence(memory_order_acquire); | |
923 | ||
03868246 JR |
924 | /* Determine 'tags' such that, if 'subtable->tag' doesn't intersect them, |
925 | * then 'flow' cannot possibly match in 'subtable': | |
c906cedf BP |
926 | * |
927 | * - If flow->metadata maps to a given 'partition', then we can use | |
928 | * 'tags' for 'partition->tags'. | |
929 | * | |
930 | * - If flow->metadata has no partition, then no rule in 'cls' has an | |
931 | * exact-match for flow->metadata. That means that we don't need to | |
03868246 | 932 | * search any subtable that includes flow->metadata in its mask. |
c906cedf | 933 | * |
03868246 | 934 | * In either case, we always need to search any cls_subtables that do not |
c906cedf | 935 | * include flow->metadata in its mask. One way to do that would be to |
03868246 JR |
936 | * check the "cls_subtable"s explicitly for that, but that would require an |
937 | * extra branch per subtable. Instead, we mark such a cls_subtable's | |
938 | * 'tags' as TAG_ALL and make sure that 'tags' is never empty. This means | |
939 | * that 'tags' always intersects such a cls_subtable's 'tags', so we don't | |
940 | * need a special case. | |
c906cedf | 941 | */ |
f2c21402 | 942 | partition = (cmap_is_empty(&cls->partitions) |
c906cedf BP |
943 | ? NULL |
944 | : find_partition(cls, flow->metadata, | |
945 | hash_metadata(flow->metadata))); | |
946 | tags = partition ? partition->tags : TAG_ARBITRARY; | |
48c3de13 | 947 | |
13751fd8 | 948 | /* Initialize trie contexts for match_find_wc(). */ |
fe7cfa5c | 949 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
950 | trie_ctx_init(&trie_ctx[i], &cls->tries[i]); |
951 | } | |
ec988646 | 952 | |
b5d97350 | 953 | best = NULL; |
fe7cfa5c JR |
954 | PVECTOR_FOR_EACH_PRIORITY(subtable, best_priority, 2, |
955 | sizeof(struct cls_subtable), &cls->subtables) { | |
627fb667 | 956 | struct cls_match *rule; |
c906cedf | 957 | |
fe7cfa5c | 958 | if (!tag_intersects(tags, subtable->tag)) { |
c906cedf BP |
959 | continue; |
960 | } | |
74f74083 | 961 | |
fe7cfa5c | 962 | rule = find_match_wc(subtable, flow, trie_ctx, cls->n_tries, wc); |
ec988646 JR |
963 | if (rule && (int64_t)rule->priority > best_priority) { |
964 | best_priority = (int64_t)rule->priority; | |
1f3c5efc | 965 | best = rule; |
b5d97350 | 966 | } |
48c3de13 | 967 | } |
13751fd8 | 968 | |
627fb667 | 969 | return best ? best->cls_rule : NULL; |
48c3de13 BP |
970 | } |
971 | ||
2abf78ff | 972 | /* Returns true if 'target' satisifies 'match', that is, if each bit for which |
3016f3e4 JR |
973 | * 'match' specifies a particular value has the correct value in 'target'. |
974 | * | |
975 | * 'flow' and 'mask' have the same mask! */ | |
2abf78ff | 976 | static bool |
3016f3e4 JR |
977 | miniflow_and_mask_matches_miniflow(const struct miniflow *flow, |
978 | const struct minimask *mask, | |
979 | const struct miniflow *target) | |
2abf78ff | 980 | { |
3016f3e4 JR |
981 | const uint32_t *flowp = miniflow_get_u32_values(flow); |
982 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
2abf78ff JR |
983 | uint32_t target_u32; |
984 | ||
3016f3e4 | 985 | MINIFLOW_FOR_EACH_IN_MAP(target_u32, target, mask->masks.map) { |
2abf78ff JR |
986 | if ((*flowp++ ^ target_u32) & *maskp++) { |
987 | return false; | |
988 | } | |
989 | } | |
990 | ||
991 | return true; | |
992 | } | |
993 | ||
b7648634 | 994 | /* For each miniflow in 'flows' performs a classifier lookup writing the result |
52a524eb | 995 | * into the corresponding slot in 'rules'. |
2abf78ff | 996 | * |
b7648634 EJ |
997 | * This function is optimized for use in the userspace datapath and therefore |
998 | * does not implement a lot of features available in the standard | |
999 | * classifier_lookup() function. Specifically, it does not implement | |
56394293 EJ |
1000 | * priorities, instead returning any rule which matches the flow. |
1001 | * | |
1002 | * Returns true if all flows found a corresponding rule. */ | |
1003 | bool | |
e48eccd1 | 1004 | classifier_lookup_miniflow_batch(const struct classifier *cls, |
52a524eb JR |
1005 | const struct miniflow *flows[], |
1006 | struct cls_rule *rules[], const size_t cnt) | |
2abf78ff | 1007 | { |
52a524eb JR |
1008 | /* The batch size 16 was experimentally found faster than 8 or 32. */ |
1009 | typedef uint16_t map_type; | |
1010 | #define MAP_BITS (sizeof(map_type) * CHAR_BIT) | |
1011 | ||
2abf78ff | 1012 | struct cls_subtable *subtable; |
52a524eb JR |
1013 | const int n_maps = DIV_ROUND_UP(cnt, MAP_BITS); |
1014 | ||
1015 | #if !defined(__CHECKER__) && !defined(_WIN32) | |
1016 | map_type maps[n_maps]; | |
1017 | #else | |
1018 | map_type maps[DIV_ROUND_UP(CLASSIFIER_MAX_BATCH, MAP_BITS)]; | |
1019 | ovs_assert(n_maps <= CLASSIFIER_MAX_BATCH); | |
1020 | #endif | |
1021 | BUILD_ASSERT_DECL(sizeof *maps * CHAR_BIT == MAP_BITS); | |
1022 | ||
1023 | memset(maps, 0xff, sizeof maps); | |
1024 | if (cnt % MAP_BITS) { | |
1025 | maps[n_maps - 1] >>= MAP_BITS - cnt % MAP_BITS; /* Clear extra bits. */ | |
1026 | } | |
1027 | memset(rules, 0, cnt * sizeof *rules); | |
2abf78ff | 1028 | |
fe7cfa5c | 1029 | PVECTOR_FOR_EACH (subtable, &cls->subtables) { |
52a524eb JR |
1030 | const struct miniflow **mfs = flows; |
1031 | struct cls_rule **results = rules; | |
1032 | map_type remains = 0; | |
1033 | int m; | |
2abf78ff | 1034 | |
52a524eb JR |
1035 | BUILD_ASSERT_DECL(sizeof remains == sizeof *maps); |
1036 | ||
1037 | for (m = 0; m < n_maps; m++, mfs += MAP_BITS, results += MAP_BITS) { | |
1038 | uint32_t hashes[MAP_BITS]; | |
1039 | const struct cmap_node *nodes[MAP_BITS]; | |
1040 | unsigned long map = maps[m]; | |
1041 | int i; | |
b7648634 | 1042 | |
52a524eb JR |
1043 | if (!map) { |
1044 | continue; /* Skip empty ones. */ | |
b7648634 | 1045 | } |
2abf78ff | 1046 | |
52a524eb JR |
1047 | /* Compute hashes for the unfound flows. */ |
1048 | ULONG_FOR_EACH_1(i, map) { | |
1049 | hashes[i] = miniflow_hash_in_minimask(mfs[i], &subtable->mask, | |
1050 | 0); | |
1051 | } | |
1052 | /* Lookup. */ | |
1053 | map = cmap_find_batch(&subtable->rules, map, hashes, nodes); | |
1054 | /* Check results. */ | |
1055 | ULONG_FOR_EACH_1(i, map) { | |
1056 | struct cls_match *rule; | |
1057 | ||
1058 | CMAP_NODE_FOR_EACH (rule, cmap_node, nodes[i]) { | |
1059 | if (OVS_LIKELY(miniflow_and_mask_matches_miniflow( | |
1060 | &rule->flow, &subtable->mask, | |
1061 | mfs[i]))) { | |
1062 | results[i] = rule->cls_rule; | |
1063 | goto next; | |
1064 | } | |
1065 | } | |
1066 | ULONG_SET0(map, i); /* Did not match. */ | |
1067 | next: | |
1068 | ; /* Keep Sparse happy. */ | |
1069 | } | |
1070 | maps[m] &= ~map; /* Clear the found rules. */ | |
1071 | remains |= maps[m]; | |
b7648634 | 1072 | } |
52a524eb JR |
1073 | if (!remains) { |
1074 | return true; /* All found. */ | |
b7648634 EJ |
1075 | } |
1076 | } | |
52a524eb | 1077 | /* Some misses. */ |
56394293 | 1078 | return false; |
2abf78ff JR |
1079 | } |
1080 | ||
b5d97350 BP |
1081 | /* Finds and returns a rule in 'cls' with exactly the same priority and |
1082 | * matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
c084ce1d | 1083 | * contain an exact match. */ |
064af421 | 1084 | struct cls_rule * |
e48eccd1 | 1085 | classifier_find_rule_exactly(const struct classifier *cls, |
76ecc721 | 1086 | const struct cls_rule *target) |
e48eccd1 | 1087 | OVS_EXCLUDED(cls->mutex) |
064af421 | 1088 | { |
627fb667 | 1089 | struct cls_match *head, *rule; |
03868246 | 1090 | struct cls_subtable *subtable; |
064af421 | 1091 | |
03868246 | 1092 | subtable = find_subtable(cls, &target->match.mask); |
d2064437 JR |
1093 | if (!subtable || target->priority > subtable->max_priority) { |
1094 | return NULL; | |
4d935a6b JR |
1095 | } |
1096 | ||
03868246 | 1097 | head = find_equal(subtable, &target->match.flow, |
5cb7a798 BP |
1098 | miniflow_hash_in_minimask(&target->match.flow, |
1099 | &target->match.mask, 0)); | |
d2064437 JR |
1100 | |
1101 | ||
1102 | /* Use RCU list instead of locking when one is available. */ | |
1103 | ovs_mutex_lock(&cls->mutex); | |
b5d97350 BP |
1104 | FOR_EACH_RULE_IN_LIST (rule, head) { |
1105 | if (target->priority >= rule->priority) { | |
e65413ab | 1106 | ovs_mutex_unlock(&cls->mutex); |
627fb667 | 1107 | return target->priority == rule->priority ? rule->cls_rule : NULL; |
064af421 BP |
1108 | } |
1109 | } | |
e65413ab | 1110 | ovs_mutex_unlock(&cls->mutex); |
d2064437 | 1111 | |
064af421 BP |
1112 | return NULL; |
1113 | } | |
1114 | ||
81a76618 BP |
1115 | /* Finds and returns a rule in 'cls' with priority 'priority' and exactly the |
1116 | * same matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
1117 | * contain an exact match. */ | |
1118 | struct cls_rule * | |
1119 | classifier_find_match_exactly(const struct classifier *cls, | |
1120 | const struct match *target, | |
1121 | unsigned int priority) | |
1122 | { | |
1123 | struct cls_rule *retval; | |
1124 | struct cls_rule cr; | |
1125 | ||
1126 | cls_rule_init(&cr, target, priority); | |
1127 | retval = classifier_find_rule_exactly(cls, &cr); | |
48d28ac1 | 1128 | cls_rule_destroy(&cr); |
81a76618 BP |
1129 | |
1130 | return retval; | |
1131 | } | |
1132 | ||
faa50f40 BP |
1133 | /* Checks if 'target' would overlap any other rule in 'cls'. Two rules are |
1134 | * considered to overlap if both rules have the same priority and a packet | |
1135 | * could match both. */ | |
49bdc010 | 1136 | bool |
e48eccd1 | 1137 | classifier_rule_overlaps(const struct classifier *cls, |
faa50f40 | 1138 | const struct cls_rule *target) |
e48eccd1 | 1139 | OVS_EXCLUDED(cls->mutex) |
49bdc010 | 1140 | { |
03868246 | 1141 | struct cls_subtable *subtable; |
fe7cfa5c | 1142 | int64_t stop_at_priority = (int64_t)target->priority - 1; |
49bdc010 | 1143 | |
e65413ab | 1144 | ovs_mutex_lock(&cls->mutex); |
03868246 | 1145 | /* Iterate subtables in the descending max priority order. */ |
fe7cfa5c JR |
1146 | PVECTOR_FOR_EACH_PRIORITY (subtable, stop_at_priority, 2, |
1147 | sizeof(struct cls_subtable), &cls->subtables) { | |
5cb7a798 BP |
1148 | uint32_t storage[FLOW_U32S]; |
1149 | struct minimask mask; | |
627fb667 | 1150 | struct cls_match *head; |
49bdc010 | 1151 | |
03868246 | 1152 | minimask_combine(&mask, &target->match.mask, &subtable->mask, storage); |
f2c21402 | 1153 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 1154 | struct cls_match *rule; |
49bdc010 | 1155 | |
b5d97350 | 1156 | FOR_EACH_RULE_IN_LIST (rule, head) { |
4d935a6b JR |
1157 | if (rule->priority < target->priority) { |
1158 | break; /* Rules in descending priority order. */ | |
1159 | } | |
faa50f40 | 1160 | if (rule->priority == target->priority |
5cb7a798 | 1161 | && miniflow_equal_in_minimask(&target->match.flow, |
3016f3e4 | 1162 | &rule->flow, &mask)) { |
e65413ab | 1163 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
1164 | return true; |
1165 | } | |
1166 | } | |
1167 | } | |
1168 | } | |
1169 | ||
e65413ab | 1170 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
1171 | return false; |
1172 | } | |
6ceeaa92 BP |
1173 | |
1174 | /* Returns true if 'rule' exactly matches 'criteria' or if 'rule' is more | |
1175 | * specific than 'criteria'. That is, 'rule' matches 'criteria' and this | |
1176 | * function returns true if, for every field: | |
1177 | * | |
1178 | * - 'criteria' and 'rule' specify the same (non-wildcarded) value for the | |
1179 | * field, or | |
1180 | * | |
1181 | * - 'criteria' wildcards the field, | |
1182 | * | |
1183 | * Conversely, 'rule' does not match 'criteria' and this function returns false | |
1184 | * if, for at least one field: | |
1185 | * | |
1186 | * - 'criteria' and 'rule' specify different values for the field, or | |
1187 | * | |
1188 | * - 'criteria' specifies a value for the field but 'rule' wildcards it. | |
1189 | * | |
1190 | * Equivalently, the truth table for whether a field matches is: | |
1191 | * | |
1192 | * rule | |
1193 | * | |
1194 | * c wildcard exact | |
1195 | * r +---------+---------+ | |
1196 | * i wild | yes | yes | | |
1197 | * t card | | | | |
1198 | * e +---------+---------+ | |
1199 | * r exact | no |if values| | |
1200 | * i | |are equal| | |
1201 | * a +---------+---------+ | |
1202 | * | |
1203 | * This is the matching rule used by OpenFlow 1.0 non-strict OFPT_FLOW_MOD | |
1204 | * commands and by OpenFlow 1.0 aggregate and flow stats. | |
1205 | * | |
81a76618 | 1206 | * Ignores rule->priority. */ |
6ceeaa92 BP |
1207 | bool |
1208 | cls_rule_is_loose_match(const struct cls_rule *rule, | |
5cb7a798 | 1209 | const struct minimatch *criteria) |
6ceeaa92 | 1210 | { |
5cb7a798 BP |
1211 | return (!minimask_has_extra(&rule->match.mask, &criteria->mask) |
1212 | && miniflow_equal_in_minimask(&rule->match.flow, &criteria->flow, | |
1213 | &criteria->mask)); | |
6ceeaa92 | 1214 | } |
b5d97350 | 1215 | \f |
5ecc9d81 BP |
1216 | /* Iteration. */ |
1217 | ||
1218 | static bool | |
627fb667 | 1219 | rule_matches(const struct cls_match *rule, const struct cls_rule *target) |
5ecc9d81 BP |
1220 | { |
1221 | return (!target | |
3016f3e4 | 1222 | || miniflow_equal_in_minimask(&rule->flow, |
5cb7a798 BP |
1223 | &target->match.flow, |
1224 | &target->match.mask)); | |
5ecc9d81 BP |
1225 | } |
1226 | ||
627fb667 | 1227 | static struct cls_match * |
03868246 | 1228 | search_subtable(const struct cls_subtable *subtable, |
f2c21402 | 1229 | struct cls_cursor *cursor) |
5ecc9d81 | 1230 | { |
f2c21402 JR |
1231 | if (!cursor->target |
1232 | || !minimask_has_extra(&subtable->mask, &cursor->target->match.mask)) { | |
627fb667 | 1233 | struct cls_match *rule; |
5ecc9d81 | 1234 | |
f2c21402 JR |
1235 | CMAP_CURSOR_FOR_EACH (rule, cmap_node, &cursor->rules, |
1236 | &subtable->rules) { | |
1237 | if (rule_matches(rule, cursor->target)) { | |
5ecc9d81 BP |
1238 | return rule; |
1239 | } | |
1240 | } | |
1241 | } | |
1242 | return NULL; | |
1243 | } | |
1244 | ||
5f0476ce JR |
1245 | /* Initializes 'cursor' for iterating through rules in 'cls', and returns the |
1246 | * first matching cls_rule via '*pnode', or NULL if there are no matches. | |
5ecc9d81 | 1247 | * |
6ceeaa92 | 1248 | * - If 'target' is null, the cursor will visit every rule in 'cls'. |
5ecc9d81 | 1249 | * |
6ceeaa92 BP |
1250 | * - If 'target' is nonnull, the cursor will visit each 'rule' in 'cls' |
1251 | * such that cls_rule_is_loose_match(rule, target) returns true. | |
5ecc9d81 | 1252 | * |
6ceeaa92 | 1253 | * Ignores target->priority. */ |
78c8df12 BP |
1254 | struct cls_cursor cls_cursor_start(const struct classifier *cls, |
1255 | const struct cls_rule *target, | |
1256 | bool safe) | |
5f0476ce | 1257 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 1258 | { |
5f0476ce | 1259 | struct cls_cursor cursor; |
03868246 | 1260 | struct cls_subtable *subtable; |
5ecc9d81 | 1261 | |
5f0476ce | 1262 | cursor.safe = safe; |
e48eccd1 | 1263 | cursor.cls = cls; |
5f0476ce | 1264 | cursor.target = target && !cls_rule_is_catchall(target) ? target : NULL; |
78c8df12 | 1265 | cursor.rule = NULL; |
5f0476ce JR |
1266 | |
1267 | /* Find first rule. */ | |
e65413ab | 1268 | ovs_mutex_lock(&cursor.cls->mutex); |
5f0476ce | 1269 | CMAP_CURSOR_FOR_EACH (subtable, cmap_node, &cursor.subtables, |
e65413ab | 1270 | &cursor.cls->subtables_map) { |
5f0476ce | 1271 | struct cls_match *rule = search_subtable(subtable, &cursor); |
f2c21402 | 1272 | |
5ecc9d81 | 1273 | if (rule) { |
5f0476ce | 1274 | cursor.subtable = subtable; |
78c8df12 | 1275 | cursor.rule = rule->cls_rule; |
5f0476ce | 1276 | break; |
5ecc9d81 BP |
1277 | } |
1278 | } | |
1279 | ||
5f0476ce | 1280 | /* Leave locked if requested and have a rule. */ |
78c8df12 | 1281 | if (safe || !cursor.rule) { |
e65413ab | 1282 | ovs_mutex_unlock(&cursor.cls->mutex); |
5f0476ce JR |
1283 | } |
1284 | return cursor; | |
1285 | } | |
1286 | ||
1caa1561 BP |
1287 | static struct cls_rule * |
1288 | cls_cursor_next(struct cls_cursor *cursor) | |
5f0476ce | 1289 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 1290 | { |
78c8df12 | 1291 | struct cls_match *rule = cursor->rule->cls_match; |
03868246 | 1292 | const struct cls_subtable *subtable; |
627fb667 | 1293 | struct cls_match *next; |
5ecc9d81 | 1294 | |
955f579d BP |
1295 | next = next_rule_in_list__(rule); |
1296 | if (next->priority < rule->priority) { | |
1caa1561 | 1297 | return next->cls_rule; |
5ecc9d81 BP |
1298 | } |
1299 | ||
955f579d | 1300 | /* 'next' is the head of the list, that is, the rule that is included in |
f2c21402 | 1301 | * the subtable's map. (This is important when the classifier contains |
03868246 | 1302 | * rules that differ only in priority.) */ |
955f579d | 1303 | rule = next; |
f2c21402 | 1304 | CMAP_CURSOR_FOR_EACH_CONTINUE (rule, cmap_node, &cursor->rules) { |
5ecc9d81 | 1305 | if (rule_matches(rule, cursor->target)) { |
1caa1561 | 1306 | return rule->cls_rule; |
5ecc9d81 BP |
1307 | } |
1308 | } | |
1309 | ||
03868246 | 1310 | subtable = cursor->subtable; |
f2c21402 JR |
1311 | CMAP_CURSOR_FOR_EACH_CONTINUE (subtable, cmap_node, &cursor->subtables) { |
1312 | rule = search_subtable(subtable, cursor); | |
5ecc9d81 | 1313 | if (rule) { |
03868246 | 1314 | cursor->subtable = subtable; |
1caa1561 | 1315 | return rule->cls_rule; |
5ecc9d81 BP |
1316 | } |
1317 | } | |
1318 | ||
1caa1561 BP |
1319 | return NULL; |
1320 | } | |
1321 | ||
1322 | /* Sets 'cursor->rule' to the next matching cls_rule in 'cursor''s iteration, | |
1323 | * or to null if all matching rules have been visited. */ | |
1324 | void | |
1325 | cls_cursor_advance(struct cls_cursor *cursor) | |
1326 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
1327 | { | |
1328 | if (cursor->safe) { | |
1329 | ovs_mutex_lock(&cursor->cls->mutex); | |
1330 | } | |
1331 | cursor->rule = cls_cursor_next(cursor); | |
1332 | if (cursor->safe || !cursor->rule) { | |
1333 | ovs_mutex_unlock(&cursor->cls->mutex); | |
1334 | } | |
5ecc9d81 BP |
1335 | } |
1336 | \f | |
03868246 | 1337 | static struct cls_subtable * |
e48eccd1 | 1338 | find_subtable(const struct classifier *cls, const struct minimask *mask) |
b5d97350 | 1339 | { |
03868246 | 1340 | struct cls_subtable *subtable; |
064af421 | 1341 | |
f2c21402 | 1342 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, minimask_hash(mask, 0), |
5a87054c | 1343 | &cls->subtables_map) { |
03868246 JR |
1344 | if (minimask_equal(mask, &subtable->mask)) { |
1345 | return subtable; | |
064af421 BP |
1346 | } |
1347 | } | |
b5d97350 | 1348 | return NULL; |
064af421 | 1349 | } |
064af421 | 1350 | |
e65413ab | 1351 | /* The new subtable will be visible to the readers only after this. */ |
03868246 | 1352 | static struct cls_subtable * |
e48eccd1 | 1353 | insert_subtable(struct classifier *cls, const struct minimask *mask) |
e65413ab | 1354 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1355 | { |
c906cedf | 1356 | uint32_t hash = minimask_hash(mask, 0); |
03868246 | 1357 | struct cls_subtable *subtable; |
476f36e8 JR |
1358 | int i, index = 0; |
1359 | struct flow_wildcards old, new; | |
1360 | uint8_t prev; | |
3016f3e4 | 1361 | int count = count_1bits(mask->masks.map); |
064af421 | 1362 | |
3016f3e4 JR |
1363 | subtable = xzalloc(sizeof *subtable - sizeof mask->masks.inline_values |
1364 | + MINIFLOW_VALUES_SIZE(count)); | |
f2c21402 | 1365 | cmap_init(&subtable->rules); |
3016f3e4 | 1366 | miniflow_clone_inline(&subtable->mask.masks, &mask->masks, count); |
476f36e8 JR |
1367 | |
1368 | /* Init indices for segmented lookup, if any. */ | |
1369 | flow_wildcards_init_catchall(&new); | |
1370 | old = new; | |
1371 | prev = 0; | |
1372 | for (i = 0; i < cls->n_flow_segments; i++) { | |
1373 | flow_wildcards_fold_minimask_range(&new, mask, prev, | |
1374 | cls->flow_segments[i]); | |
1375 | /* Add an index if it adds mask bits. */ | |
1376 | if (!flow_wildcards_equal(&new, &old)) { | |
f2c21402 | 1377 | cmap_init(&subtable->indices[index]); |
476f36e8 JR |
1378 | subtable->index_ofs[index] = cls->flow_segments[i]; |
1379 | index++; | |
1380 | old = new; | |
1381 | } | |
1382 | prev = cls->flow_segments[i]; | |
1383 | } | |
1384 | /* Check if the rest of the subtable's mask adds any bits, | |
1385 | * and remove the last index if it doesn't. */ | |
1386 | if (index > 0) { | |
1387 | flow_wildcards_fold_minimask_range(&new, mask, prev, FLOW_U32S); | |
1388 | if (flow_wildcards_equal(&new, &old)) { | |
1389 | --index; | |
1390 | subtable->index_ofs[index] = 0; | |
f2c21402 | 1391 | cmap_destroy(&subtable->indices[index]); |
476f36e8 JR |
1392 | } |
1393 | } | |
1394 | subtable->n_indices = index; | |
1395 | ||
03868246 JR |
1396 | subtable->tag = (minimask_get_metadata_mask(mask) == OVS_BE64_MAX |
1397 | ? tag_create_deterministic(hash) | |
1398 | : TAG_ALL); | |
064af421 | 1399 | |
13751fd8 JR |
1400 | for (i = 0; i < cls->n_tries; i++) { |
1401 | subtable->trie_plen[i] = minimask_get_prefix_len(mask, | |
1402 | cls->tries[i].field); | |
1403 | } | |
1404 | ||
69d6040e | 1405 | /* Ports trie. */ |
f358a2cb | 1406 | ovsrcu_set_hidden(&subtable->ports_trie, NULL); |
69d6040e JR |
1407 | subtable->ports_mask_len |
1408 | = 32 - ctz32(ntohl(MINIFLOW_GET_BE32(&mask->masks, tp_src))); | |
1409 | ||
f2c21402 | 1410 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, hash); |
ec988646 | 1411 | |
03868246 | 1412 | return subtable; |
064af421 BP |
1413 | } |
1414 | ||
b5d97350 | 1415 | static void |
e48eccd1 | 1416 | destroy_subtable(struct classifier *cls, struct cls_subtable *subtable) |
e65413ab | 1417 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1418 | { |
476f36e8 JR |
1419 | int i; |
1420 | ||
fe7cfa5c | 1421 | pvector_remove(&cls->subtables, subtable); |
f358a2cb | 1422 | trie_destroy(&subtable->ports_trie); |
69d6040e | 1423 | |
476f36e8 | 1424 | for (i = 0; i < subtable->n_indices; i++) { |
f2c21402 | 1425 | cmap_destroy(&subtable->indices[i]); |
476f36e8 | 1426 | } |
f2c21402 JR |
1427 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, |
1428 | minimask_hash(&subtable->mask, 0)); | |
fe7cfa5c | 1429 | minimask_destroy(&subtable->mask); |
f2c21402 | 1430 | cmap_destroy(&subtable->rules); |
fe7cfa5c | 1431 | ovsrcu_postpone(free, subtable); |
4aacd02d BP |
1432 | } |
1433 | ||
13751fd8 JR |
1434 | struct range { |
1435 | uint8_t start; | |
1436 | uint8_t end; | |
1437 | }; | |
1438 | ||
c0bfb650 JR |
1439 | static unsigned int be_get_bit_at(const ovs_be32 value[], unsigned int ofs); |
1440 | ||
13751fd8 JR |
1441 | /* Return 'true' if can skip rest of the subtable based on the prefix trie |
1442 | * lookup results. */ | |
1443 | static inline bool | |
1444 | check_tries(struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, | |
1445 | const unsigned int field_plen[CLS_MAX_TRIES], | |
1446 | const struct range ofs, const struct flow *flow, | |
1447 | struct flow_wildcards *wc) | |
1448 | { | |
1449 | int j; | |
1450 | ||
1451 | /* Check if we could avoid fully unwildcarding the next level of | |
1452 | * fields using the prefix tries. The trie checks are done only as | |
1453 | * needed to avoid folding in additional bits to the wildcards mask. */ | |
1454 | for (j = 0; j < n_tries; j++) { | |
1455 | /* Is the trie field relevant for this subtable? */ | |
1456 | if (field_plen[j]) { | |
1457 | struct trie_ctx *ctx = &trie_ctx[j]; | |
1458 | uint8_t be32ofs = ctx->be32ofs; | |
1459 | ||
1460 | /* Is the trie field within the current range of fields? */ | |
1461 | if (be32ofs >= ofs.start && be32ofs < ofs.end) { | |
1462 | /* On-demand trie lookup. */ | |
1463 | if (!ctx->lookup_done) { | |
c0bfb650 JR |
1464 | memset(&ctx->match_plens, 0, sizeof ctx->match_plens); |
1465 | ctx->maskbits = trie_lookup(ctx->trie, flow, | |
1466 | &ctx->match_plens); | |
13751fd8 JR |
1467 | ctx->lookup_done = true; |
1468 | } | |
1469 | /* Possible to skip the rest of the subtable if subtable's | |
c0bfb650 JR |
1470 | * prefix on the field is not included in the lookup result. */ |
1471 | if (!be_get_bit_at(&ctx->match_plens.be32, field_plen[j] - 1)) { | |
1817dcea JR |
1472 | /* We want the trie lookup to never result in unwildcarding |
1473 | * any bits that would not be unwildcarded otherwise. | |
1474 | * Since the trie is shared by the whole classifier, it is | |
1475 | * possible that the 'maskbits' contain bits that are | |
1476 | * irrelevant for the partition relevant for the current | |
1477 | * packet. Hence the checks below. */ | |
13751fd8 | 1478 | |
13751fd8 | 1479 | /* Check that the trie result will not unwildcard more bits |
1817dcea | 1480 | * than this subtable would otherwise. */ |
13751fd8 JR |
1481 | if (ctx->maskbits <= field_plen[j]) { |
1482 | /* Unwildcard the bits and skip the rest. */ | |
1483 | mask_set_prefix_bits(wc, be32ofs, ctx->maskbits); | |
1484 | /* Note: Prerequisite already unwildcarded, as the only | |
1485 | * prerequisite of the supported trie lookup fields is | |
1817dcea JR |
1486 | * the ethertype, which is always unwildcarded. */ |
1487 | return true; | |
1488 | } | |
1489 | /* Can skip if the field is already unwildcarded. */ | |
1490 | if (mask_prefix_bits_set(wc, be32ofs, ctx->maskbits)) { | |
13751fd8 JR |
1491 | return true; |
1492 | } | |
1493 | } | |
1494 | } | |
1495 | } | |
1496 | } | |
1497 | return false; | |
1498 | } | |
1499 | ||
3016f3e4 JR |
1500 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1501 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1502 | * value has the correct value in 'target'. | |
1503 | * | |
1504 | * This function is equivalent to miniflow_equal_flow_in_minimask(flow, | |
a64759f0 JR |
1505 | * target, mask) but this is faster because of the invariant that |
1506 | * flow->map and mask->masks.map are the same, and that this version | |
1507 | * takes the 'wc'. */ | |
3016f3e4 JR |
1508 | static inline bool |
1509 | miniflow_and_mask_matches_flow(const struct miniflow *flow, | |
1510 | const struct minimask *mask, | |
e9319757 | 1511 | const struct flow *target) |
3016f3e4 JR |
1512 | { |
1513 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1514 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
a64759f0 | 1515 | uint32_t idx; |
3016f3e4 | 1516 | |
a64759f0 JR |
1517 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { |
1518 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & *maskp++; | |
1519 | ||
1520 | if (diff) { | |
3016f3e4 JR |
1521 | return false; |
1522 | } | |
1523 | } | |
1524 | ||
1525 | return true; | |
1526 | } | |
1527 | ||
627fb667 | 1528 | static inline struct cls_match * |
476f36e8 JR |
1529 | find_match(const struct cls_subtable *subtable, const struct flow *flow, |
1530 | uint32_t hash) | |
b5d97350 | 1531 | { |
627fb667 | 1532 | struct cls_match *rule; |
b5d97350 | 1533 | |
f2c21402 | 1534 | CMAP_FOR_EACH_WITH_HASH (rule, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1535 | if (miniflow_and_mask_matches_flow(&rule->flow, &subtable->mask, |
e9319757 | 1536 | flow)) { |
b5d97350 | 1537 | return rule; |
064af421 BP |
1538 | } |
1539 | } | |
c23740be | 1540 | |
064af421 BP |
1541 | return NULL; |
1542 | } | |
1543 | ||
e9319757 JR |
1544 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1545 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1546 | * value has the correct value in 'target'. | |
1547 | * | |
1548 | * This function is equivalent to miniflow_and_mask_matches_flow() but this | |
1549 | * version fills in the mask bits in 'wc'. */ | |
1550 | static inline bool | |
1551 | miniflow_and_mask_matches_flow_wc(const struct miniflow *flow, | |
1552 | const struct minimask *mask, | |
1553 | const struct flow *target, | |
1554 | struct flow_wildcards *wc) | |
1555 | { | |
1556 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1557 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
1558 | uint32_t idx; | |
1559 | ||
1560 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { | |
1561 | uint32_t mask = *maskp++; | |
1562 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & mask; | |
1563 | ||
1564 | if (diff) { | |
1565 | /* Only unwildcard if none of the differing bits is already | |
1566 | * exact-matched. */ | |
1567 | if (!(flow_u32_value(&wc->masks, idx) & diff)) { | |
1568 | /* Keep one bit of the difference. */ | |
1569 | *flow_u32_lvalue(&wc->masks, idx) |= rightmost_1bit(diff); | |
1570 | } | |
1571 | return false; | |
1572 | } | |
1573 | /* Fill in the bits that were looked at. */ | |
1574 | *flow_u32_lvalue(&wc->masks, idx) |= mask; | |
1575 | } | |
1576 | ||
1577 | return true; | |
1578 | } | |
1579 | ||
386cb9f7 JR |
1580 | /* Unwildcard the fields looked up so far, if any. */ |
1581 | static void | |
1582 | fill_range_wc(const struct cls_subtable *subtable, struct flow_wildcards *wc, | |
1583 | uint8_t to) | |
1584 | { | |
1585 | if (to) { | |
1586 | flow_wildcards_fold_minimask_range(wc, &subtable->mask, 0, to); | |
1587 | } | |
1588 | } | |
1589 | ||
627fb667 | 1590 | static struct cls_match * |
476f36e8 | 1591 | find_match_wc(const struct cls_subtable *subtable, const struct flow *flow, |
13751fd8 JR |
1592 | struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, |
1593 | struct flow_wildcards *wc) | |
476f36e8 JR |
1594 | { |
1595 | uint32_t basis = 0, hash; | |
f17e8ad6 | 1596 | struct cls_match *rule = NULL; |
476f36e8 | 1597 | int i; |
13751fd8 | 1598 | struct range ofs; |
476f36e8 | 1599 | |
ec988646 | 1600 | if (OVS_UNLIKELY(!wc)) { |
476f36e8 JR |
1601 | return find_match(subtable, flow, |
1602 | flow_hash_in_minimask(flow, &subtable->mask, 0)); | |
1603 | } | |
1604 | ||
13751fd8 | 1605 | ofs.start = 0; |
476f36e8 JR |
1606 | /* Try to finish early by checking fields in segments. */ |
1607 | for (i = 0; i < subtable->n_indices; i++) { | |
55847abe | 1608 | const struct cmap_node *inode; |
f2c21402 | 1609 | |
13751fd8 | 1610 | ofs.end = subtable->index_ofs[i]; |
476f36e8 | 1611 | |
13751fd8 JR |
1612 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, |
1613 | wc)) { | |
386cb9f7 JR |
1614 | /* 'wc' bits for the trie field set, now unwildcard the preceding |
1615 | * bits used so far. */ | |
1616 | fill_range_wc(subtable, wc, ofs.start); | |
1617 | return NULL; | |
13751fd8 JR |
1618 | } |
1619 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, | |
1620 | ofs.end, &basis); | |
f2c21402 | 1621 | inode = cmap_find(&subtable->indices[i], hash); |
476f36e8 | 1622 | if (!inode) { |
386cb9f7 JR |
1623 | /* No match, can stop immediately, but must fold in the bits |
1624 | * used in lookup so far. */ | |
1625 | fill_range_wc(subtable, wc, ofs.end); | |
1626 | return NULL; | |
476f36e8 JR |
1627 | } |
1628 | ||
1629 | /* If we have narrowed down to a single rule already, check whether | |
a64759f0 | 1630 | * that rule matches. Either way, we're done. |
476f36e8 JR |
1631 | * |
1632 | * (Rare) hash collisions may cause us to miss the opportunity for this | |
1633 | * optimization. */ | |
f2c21402 | 1634 | if (!cmap_node_next(inode)) { |
476f36e8 | 1635 | ASSIGN_CONTAINER(rule, inode - i, index_nodes); |
e9319757 JR |
1636 | if (miniflow_and_mask_matches_flow_wc(&rule->flow, &subtable->mask, |
1637 | flow, wc)) { | |
1638 | return rule; | |
476f36e8 | 1639 | } |
e9319757 | 1640 | return NULL; |
476f36e8 | 1641 | } |
386cb9f7 | 1642 | ofs.start = ofs.end; |
476f36e8 | 1643 | } |
13751fd8 JR |
1644 | ofs.end = FLOW_U32S; |
1645 | /* Trie check for the final range. */ | |
1646 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, wc)) { | |
386cb9f7 JR |
1647 | fill_range_wc(subtable, wc, ofs.start); |
1648 | return NULL; | |
13751fd8 | 1649 | } |
a64759f0 JR |
1650 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, |
1651 | ofs.end, &basis); | |
1652 | rule = find_match(subtable, flow, hash); | |
69d6040e JR |
1653 | if (!rule && subtable->ports_mask_len) { |
1654 | /* Ports are always part of the final range, if any. | |
1655 | * No match was found for the ports. Use the ports trie to figure out | |
1656 | * which ports bits to unwildcard. */ | |
1657 | unsigned int mbits; | |
c0bfb650 | 1658 | ovs_be32 value, plens, mask; |
69d6040e JR |
1659 | |
1660 | mask = MINIFLOW_GET_BE32(&subtable->mask.masks, tp_src); | |
1661 | value = ((OVS_FORCE ovs_be32 *)flow)[TP_PORTS_OFS32] & mask; | |
c0bfb650 | 1662 | mbits = trie_lookup_value(&subtable->ports_trie, &value, &plens, 32); |
69d6040e JR |
1663 | |
1664 | ((OVS_FORCE ovs_be32 *)&wc->masks)[TP_PORTS_OFS32] |= | |
1665 | mask & htonl(~0 << (32 - mbits)); | |
1666 | ||
386cb9f7 JR |
1667 | /* Unwildcard all bits in the mask upto the ports, as they were used |
1668 | * to determine there is no match. */ | |
1669 | fill_range_wc(subtable, wc, TP_PORTS_OFS32); | |
1670 | return NULL; | |
69d6040e | 1671 | } |
e9319757 | 1672 | |
13751fd8 | 1673 | /* Must unwildcard all the fields, as they were looked at. */ |
476f36e8 JR |
1674 | flow_wildcards_fold_minimask(wc, &subtable->mask); |
1675 | return rule; | |
1676 | } | |
1677 | ||
627fb667 | 1678 | static struct cls_match * |
03868246 JR |
1679 | find_equal(struct cls_subtable *subtable, const struct miniflow *flow, |
1680 | uint32_t hash) | |
064af421 | 1681 | { |
627fb667 | 1682 | struct cls_match *head; |
064af421 | 1683 | |
f2c21402 | 1684 | CMAP_FOR_EACH_WITH_HASH (head, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1685 | if (miniflow_equal(&head->flow, flow)) { |
b5d97350 | 1686 | return head; |
064af421 BP |
1687 | } |
1688 | } | |
1689 | return NULL; | |
1690 | } | |
1691 | ||
afae68b1 JR |
1692 | /* |
1693 | * As the readers are operating concurrently with the modifications, a | |
1694 | * concurrent reader may or may not see the new rule, depending on how | |
1695 | * the concurrent events overlap with each other. This is no | |
1696 | * different from the former locked behavior, but there the visibility | |
1697 | * of the new rule only depended on the timing of the locking | |
1698 | * functions. | |
1699 | * | |
1700 | * The new rule is first added to the segment indices, so the readers | |
1701 | * may find the rule in the indices before the rule is visible in the | |
1702 | * subtables 'rules' map. This may result in us losing the | |
1703 | * opportunity to quit lookups earlier, resulting in sub-optimal | |
1704 | * wildcarding. This will be fixed by forthcoming revalidation always | |
1705 | * scheduled after flow table changes. | |
1706 | * | |
1707 | * Similar behavior may happen due to us removing the overlapping rule | |
1708 | * (if any) from the indices only after the new rule has been added. | |
1709 | * | |
1710 | * The subtable's max priority is updated only after the rule is | |
1711 | * inserted, so the concurrent readers may not see the rule, as the | |
1712 | * updated priority ordered subtable list will only be visible after | |
1713 | * the subtable's max priority is updated. | |
1714 | * | |
1715 | * Similarly, the classifier's partitions for new rules are updated by | |
1716 | * the caller after this function, so the readers may keep skipping | |
1717 | * the subtable until they see the updated partitions. | |
1718 | */ | |
627fb667 | 1719 | static struct cls_match * |
e48eccd1 | 1720 | insert_rule(struct classifier *cls, struct cls_subtable *subtable, |
f2c21402 | 1721 | struct cls_rule *new_rule) |
e65413ab | 1722 | OVS_REQUIRES(cls->mutex) |
064af421 | 1723 | { |
627fb667 | 1724 | struct cls_match *old = NULL; |
f2c21402 JR |
1725 | struct cls_match *new = cls_match_alloc(new_rule); |
1726 | struct cls_match *head; | |
476f36e8 | 1727 | int i; |
f2c21402 | 1728 | uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES]; |
13751fd8 | 1729 | uint8_t prev_be32ofs = 0; |
476f36e8 JR |
1730 | |
1731 | /* Add new node to segment indices. */ | |
1732 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 JR |
1733 | ihash[i] = minimatch_hash_range(&new_rule->match, prev_be32ofs, |
1734 | subtable->index_ofs[i], &basis); | |
1735 | cmap_insert(&subtable->indices[i], &new->index_nodes[i], ihash[i]); | |
13751fd8 | 1736 | prev_be32ofs = subtable->index_ofs[i]; |
476f36e8 | 1737 | } |
f2c21402 JR |
1738 | hash = minimatch_hash_range(&new_rule->match, prev_be32ofs, FLOW_U32S, |
1739 | &basis); | |
1740 | head = find_equal(subtable, &new_rule->match.flow, hash); | |
b5d97350 | 1741 | if (!head) { |
f2c21402 JR |
1742 | cmap_insert(&subtable->rules, &new->cmap_node, hash); |
1743 | list_init(&new->list); | |
4aacd02d | 1744 | goto out; |
b5d97350 BP |
1745 | } else { |
1746 | /* Scan the list for the insertion point that will keep the list in | |
1747 | * order of decreasing priority. */ | |
627fb667 | 1748 | struct cls_match *rule; |
476f36e8 | 1749 | |
b5d97350 | 1750 | FOR_EACH_RULE_IN_LIST (rule, head) { |
f2c21402 | 1751 | if (new->priority >= rule->priority) { |
b5d97350 | 1752 | if (rule == head) { |
f2c21402 JR |
1753 | /* 'new' is the new highest-priority flow in the list. */ |
1754 | cmap_replace(&subtable->rules, &rule->cmap_node, | |
1755 | &new->cmap_node, hash); | |
b5d97350 | 1756 | } |
064af421 | 1757 | |
f2c21402 JR |
1758 | if (new->priority == rule->priority) { |
1759 | list_replace(&new->list, &rule->list); | |
4aacd02d | 1760 | old = rule; |
b5d97350 | 1761 | } else { |
f2c21402 | 1762 | list_insert(&rule->list, &new->list); |
b5d97350 | 1763 | } |
fe7cfa5c | 1764 | goto out; |
b5d97350 BP |
1765 | } |
1766 | } | |
064af421 | 1767 | |
b5d97350 | 1768 | /* Insert 'new' at the end of the list. */ |
f2c21402 | 1769 | list_push_back(&head->list, &new->list); |
064af421 | 1770 | } |
4aacd02d BP |
1771 | |
1772 | out: | |
1773 | if (!old) { | |
fe7cfa5c JR |
1774 | subtable->n_rules++; |
1775 | ||
1776 | /* Rule was added, not replaced. Update 'subtable's 'max_priority' | |
1777 | * and 'max_count', if necessary. */ | |
1778 | if (subtable->n_rules == 1) { | |
f2c21402 | 1779 | subtable->max_priority = new->priority; |
fe7cfa5c | 1780 | subtable->max_count = 1; |
f2c21402 JR |
1781 | pvector_insert(&cls->subtables, subtable, new->priority); |
1782 | } else if (subtable->max_priority == new->priority) { | |
fe7cfa5c | 1783 | ++subtable->max_count; |
f2c21402 JR |
1784 | } else if (new->priority > subtable->max_priority) { |
1785 | subtable->max_priority = new->priority; | |
fe7cfa5c | 1786 | subtable->max_count = 1; |
f2c21402 | 1787 | pvector_change_priority(&cls->subtables, subtable, new->priority); |
fe7cfa5c | 1788 | } |
476f36e8 JR |
1789 | } else { |
1790 | /* Remove old node from indices. */ | |
1791 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 | 1792 | cmap_remove(&subtable->indices[i], &old->index_nodes[i], ihash[i]); |
476f36e8 | 1793 | } |
4aacd02d BP |
1794 | } |
1795 | return old; | |
064af421 BP |
1796 | } |
1797 | ||
627fb667 JR |
1798 | static struct cls_match * |
1799 | next_rule_in_list__(struct cls_match *rule) | |
e65413ab | 1800 | OVS_NO_THREAD_SAFETY_ANALYSIS |
064af421 | 1801 | { |
f17e8ad6 GS |
1802 | struct cls_match *next = NULL; |
1803 | next = OBJECT_CONTAINING(rule->list.next, next, list); | |
955f579d BP |
1804 | return next; |
1805 | } | |
1806 | ||
627fb667 JR |
1807 | static struct cls_match * |
1808 | next_rule_in_list(struct cls_match *rule) | |
955f579d | 1809 | { |
627fb667 | 1810 | struct cls_match *next = next_rule_in_list__(rule); |
b5d97350 | 1811 | return next->priority < rule->priority ? next : NULL; |
064af421 | 1812 | } |
13751fd8 JR |
1813 | \f |
1814 | /* A longest-prefix match tree. */ | |
1815 | struct trie_node { | |
1816 | uint32_t prefix; /* Prefix bits for this node, MSB first. */ | |
c30cfa6b | 1817 | uint8_t n_bits; /* Never zero, except for the root node. */ |
13751fd8 | 1818 | unsigned int n_rules; /* Number of rules that have this prefix. */ |
f358a2cb | 1819 | rcu_trie_ptr edges[2]; /* Both NULL if leaf. */ |
13751fd8 JR |
1820 | }; |
1821 | ||
1822 | /* Max bits per node. Must fit in struct trie_node's 'prefix'. | |
1823 | * Also tested with 16, 8, and 5 to stress the implementation. */ | |
1824 | #define TRIE_PREFIX_BITS 32 | |
1825 | ||
1826 | /* Return at least 'plen' bits of the 'prefix', starting at bit offset 'ofs'. | |
1827 | * Prefixes are in the network byte order, and the offset 0 corresponds to | |
1828 | * the most significant bit of the first byte. The offset can be read as | |
1829 | * "how many bits to skip from the start of the prefix starting at 'pr'". */ | |
1830 | static uint32_t | |
1831 | raw_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1832 | { | |
1833 | uint32_t prefix; | |
1834 | ||
1835 | pr += ofs / 32; /* Where to start. */ | |
1836 | ofs %= 32; /* How many bits to skip at 'pr'. */ | |
1837 | ||
1838 | prefix = ntohl(*pr) << ofs; /* Get the first 32 - ofs bits. */ | |
1839 | if (plen > 32 - ofs) { /* Need more than we have already? */ | |
1840 | prefix |= ntohl(*++pr) >> (32 - ofs); | |
1841 | } | |
1842 | /* Return with possible unwanted bits at the end. */ | |
1843 | return prefix; | |
1844 | } | |
1845 | ||
1846 | /* Return min(TRIE_PREFIX_BITS, plen) bits of the 'prefix', starting at bit | |
1847 | * offset 'ofs'. Prefixes are in the network byte order, and the offset 0 | |
1848 | * corresponds to the most significant bit of the first byte. The offset can | |
1849 | * be read as "how many bits to skip from the start of the prefix starting at | |
1850 | * 'pr'". */ | |
1851 | static uint32_t | |
1852 | trie_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1853 | { | |
1854 | if (!plen) { | |
1855 | return 0; | |
1856 | } | |
1857 | if (plen > TRIE_PREFIX_BITS) { | |
1858 | plen = TRIE_PREFIX_BITS; /* Get at most TRIE_PREFIX_BITS. */ | |
1859 | } | |
1860 | /* Return with unwanted bits cleared. */ | |
1861 | return raw_get_prefix(pr, ofs, plen) & ~0u << (32 - plen); | |
1862 | } | |
1863 | ||
c30cfa6b | 1864 | /* Return the number of equal bits in 'n_bits' of 'prefix's MSBs and a 'value' |
13751fd8 JR |
1865 | * starting at "MSB 0"-based offset 'ofs'. */ |
1866 | static unsigned int | |
c30cfa6b | 1867 | prefix_equal_bits(uint32_t prefix, unsigned int n_bits, const ovs_be32 value[], |
13751fd8 JR |
1868 | unsigned int ofs) |
1869 | { | |
c30cfa6b | 1870 | uint64_t diff = prefix ^ raw_get_prefix(value, ofs, n_bits); |
13751fd8 | 1871 | /* Set the bit after the relevant bits to limit the result. */ |
c30cfa6b | 1872 | return raw_clz64(diff << 32 | UINT64_C(1) << (63 - n_bits)); |
13751fd8 JR |
1873 | } |
1874 | ||
1875 | /* Return the number of equal bits in 'node' prefix and a 'prefix' of length | |
1876 | * 'plen', starting at "MSB 0"-based offset 'ofs'. */ | |
1877 | static unsigned int | |
1878 | trie_prefix_equal_bits(const struct trie_node *node, const ovs_be32 prefix[], | |
1879 | unsigned int ofs, unsigned int plen) | |
1880 | { | |
c30cfa6b | 1881 | return prefix_equal_bits(node->prefix, MIN(node->n_bits, plen - ofs), |
13751fd8 JR |
1882 | prefix, ofs); |
1883 | } | |
1884 | ||
1885 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' can | |
1886 | * be greater than 31. */ | |
1887 | static unsigned int | |
1888 | be_get_bit_at(const ovs_be32 value[], unsigned int ofs) | |
1889 | { | |
1890 | return (((const uint8_t *)value)[ofs / 8] >> (7 - ofs % 8)) & 1u; | |
1891 | } | |
1892 | ||
1893 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' must | |
1894 | * be between 0 and 31, inclusive. */ | |
1895 | static unsigned int | |
1896 | get_bit_at(const uint32_t prefix, unsigned int ofs) | |
1897 | { | |
1898 | return (prefix >> (31 - ofs)) & 1u; | |
1899 | } | |
1900 | ||
1901 | /* Create new branch. */ | |
1902 | static struct trie_node * | |
1903 | trie_branch_create(const ovs_be32 *prefix, unsigned int ofs, unsigned int plen, | |
1904 | unsigned int n_rules) | |
1905 | { | |
1906 | struct trie_node *node = xmalloc(sizeof *node); | |
1907 | ||
1908 | node->prefix = trie_get_prefix(prefix, ofs, plen); | |
1909 | ||
1910 | if (plen <= TRIE_PREFIX_BITS) { | |
c30cfa6b | 1911 | node->n_bits = plen; |
f358a2cb JR |
1912 | ovsrcu_set_hidden(&node->edges[0], NULL); |
1913 | ovsrcu_set_hidden(&node->edges[1], NULL); | |
13751fd8 JR |
1914 | node->n_rules = n_rules; |
1915 | } else { /* Need intermediate nodes. */ | |
1916 | struct trie_node *subnode = trie_branch_create(prefix, | |
1917 | ofs + TRIE_PREFIX_BITS, | |
1918 | plen - TRIE_PREFIX_BITS, | |
1919 | n_rules); | |
1920 | int bit = get_bit_at(subnode->prefix, 0); | |
c30cfa6b | 1921 | node->n_bits = TRIE_PREFIX_BITS; |
f358a2cb JR |
1922 | ovsrcu_set_hidden(&node->edges[bit], subnode); |
1923 | ovsrcu_set_hidden(&node->edges[!bit], NULL); | |
13751fd8 JR |
1924 | node->n_rules = 0; |
1925 | } | |
1926 | return node; | |
1927 | } | |
1928 | ||
1929 | static void | |
f358a2cb | 1930 | trie_node_destroy(const struct trie_node *node) |
13751fd8 | 1931 | { |
f358a2cb JR |
1932 | ovsrcu_postpone(free, CONST_CAST(struct trie_node *, node)); |
1933 | } | |
1934 | ||
1935 | /* Copy a trie node for modification and postpone delete the old one. */ | |
1936 | static struct trie_node * | |
1937 | trie_node_rcu_realloc(const struct trie_node *node) | |
1938 | { | |
1939 | struct trie_node *new_node = xmalloc(sizeof *node); | |
1940 | ||
1941 | *new_node = *node; | |
1942 | trie_node_destroy(node); | |
1943 | ||
1944 | return new_node; | |
13751fd8 JR |
1945 | } |
1946 | ||
e48eccd1 | 1947 | /* May only be called while holding the classifier mutex. */ |
13751fd8 | 1948 | static void |
f358a2cb | 1949 | trie_destroy(rcu_trie_ptr *trie) |
13751fd8 | 1950 | { |
f358a2cb JR |
1951 | struct trie_node *node = ovsrcu_get_protected(struct trie_node *, trie); |
1952 | ||
13751fd8 | 1953 | if (node) { |
f358a2cb JR |
1954 | ovsrcu_set_hidden(trie, NULL); |
1955 | trie_destroy(&node->edges[0]); | |
1956 | trie_destroy(&node->edges[1]); | |
1957 | trie_node_destroy(node); | |
13751fd8 JR |
1958 | } |
1959 | } | |
1960 | ||
1961 | static bool | |
1962 | trie_is_leaf(const struct trie_node *trie) | |
1963 | { | |
f358a2cb JR |
1964 | /* No children? */ |
1965 | return !ovsrcu_get(struct trie_node *, &trie->edges[0]) | |
1966 | && !ovsrcu_get(struct trie_node *, &trie->edges[1]); | |
13751fd8 JR |
1967 | } |
1968 | ||
1969 | static void | |
1970 | mask_set_prefix_bits(struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1971 | unsigned int n_bits) |
13751fd8 JR |
1972 | { |
1973 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1974 | unsigned int i; | |
1975 | ||
c30cfa6b | 1976 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1977 | mask[i] = OVS_BE32_MAX; |
1978 | } | |
c30cfa6b JR |
1979 | if (n_bits % 32) { |
1980 | mask[i] |= htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1981 | } |
1982 | } | |
1983 | ||
1984 | static bool | |
1985 | mask_prefix_bits_set(const struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1986 | unsigned int n_bits) |
13751fd8 JR |
1987 | { |
1988 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1989 | unsigned int i; | |
1990 | ovs_be32 zeroes = 0; | |
1991 | ||
c30cfa6b | 1992 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1993 | zeroes |= ~mask[i]; |
1994 | } | |
c30cfa6b JR |
1995 | if (n_bits % 32) { |
1996 | zeroes |= ~mask[i] & htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1997 | } |
1998 | ||
c30cfa6b | 1999 | return !zeroes; /* All 'n_bits' bits set. */ |
13751fd8 JR |
2000 | } |
2001 | ||
f358a2cb | 2002 | static rcu_trie_ptr * |
13751fd8 JR |
2003 | trie_next_edge(struct trie_node *node, const ovs_be32 value[], |
2004 | unsigned int ofs) | |
2005 | { | |
2006 | return node->edges + be_get_bit_at(value, ofs); | |
2007 | } | |
2008 | ||
2009 | static const struct trie_node * | |
2010 | trie_next_node(const struct trie_node *node, const ovs_be32 value[], | |
2011 | unsigned int ofs) | |
2012 | { | |
f358a2cb JR |
2013 | return ovsrcu_get(struct trie_node *, |
2014 | &node->edges[be_get_bit_at(value, ofs)]); | |
13751fd8 JR |
2015 | } |
2016 | ||
c0bfb650 JR |
2017 | /* Set the bit at ("MSB 0"-based) offset 'ofs'. 'ofs' can be greater than 31. |
2018 | */ | |
2019 | static void | |
2020 | be_set_bit_at(ovs_be32 value[], unsigned int ofs) | |
2021 | { | |
2022 | ((uint8_t *)value)[ofs / 8] |= 1u << (7 - ofs % 8); | |
2023 | } | |
2024 | ||
2025 | /* Returns the number of bits in the prefix mask necessary to determine a | |
2026 | * mismatch, in case there are longer prefixes in the tree below the one that | |
2027 | * matched. | |
2028 | * '*plens' will have a bit set for each prefix length that may have matching | |
2029 | * rules. The caller is responsible for clearing the '*plens' prior to | |
2030 | * calling this. | |
13751fd8 JR |
2031 | */ |
2032 | static unsigned int | |
f358a2cb | 2033 | trie_lookup_value(const rcu_trie_ptr *trie, const ovs_be32 value[], |
c0bfb650 | 2034 | ovs_be32 plens[], unsigned int n_bits) |
13751fd8 | 2035 | { |
13751fd8 | 2036 | const struct trie_node *prev = NULL; |
c0bfb650 JR |
2037 | const struct trie_node *node = ovsrcu_get(struct trie_node *, trie); |
2038 | unsigned int match_len = 0; /* Number of matching bits. */ | |
13751fd8 | 2039 | |
27ce650f | 2040 | for (; node; prev = node, node = trie_next_node(node, value, match_len)) { |
13751fd8 JR |
2041 | unsigned int eqbits; |
2042 | /* Check if this edge can be followed. */ | |
27ce650f JR |
2043 | eqbits = prefix_equal_bits(node->prefix, node->n_bits, value, |
2044 | match_len); | |
2045 | match_len += eqbits; | |
c30cfa6b | 2046 | if (eqbits < node->n_bits) { /* Mismatch, nothing more to be found. */ |
27ce650f | 2047 | /* Bit at offset 'match_len' differed. */ |
c0bfb650 | 2048 | return match_len + 1; /* Includes the first mismatching bit. */ |
13751fd8 JR |
2049 | } |
2050 | /* Full match, check if rules exist at this prefix length. */ | |
2051 | if (node->n_rules > 0) { | |
c0bfb650 | 2052 | be_set_bit_at(plens, match_len - 1); |
13751fd8 | 2053 | } |
27ce650f | 2054 | if (match_len >= n_bits) { |
c0bfb650 | 2055 | return n_bits; /* Full prefix. */ |
f0e5aa11 | 2056 | } |
13751fd8 | 2057 | } |
c0bfb650 JR |
2058 | /* node == NULL. Full match so far, but we tried to follow an |
2059 | * non-existing branch. Need to exclude the other branch if it exists | |
2060 | * (it does not if we were called on an empty trie or 'prev' is a leaf | |
2061 | * node). */ | |
2062 | return !prev || trie_is_leaf(prev) ? match_len : match_len + 1; | |
13751fd8 JR |
2063 | } |
2064 | ||
2065 | static unsigned int | |
2066 | trie_lookup(const struct cls_trie *trie, const struct flow *flow, | |
c0bfb650 | 2067 | union mf_value *plens) |
13751fd8 JR |
2068 | { |
2069 | const struct mf_field *mf = trie->field; | |
2070 | ||
2071 | /* Check that current flow matches the prerequisites for the trie | |
2072 | * field. Some match fields are used for multiple purposes, so we | |
2073 | * must check that the trie is relevant for this flow. */ | |
2074 | if (mf_are_prereqs_ok(mf, flow)) { | |
f358a2cb | 2075 | return trie_lookup_value(&trie->root, |
13751fd8 | 2076 | &((ovs_be32 *)flow)[mf->flow_be32ofs], |
c0bfb650 | 2077 | &plens->be32, mf->n_bits); |
13751fd8 | 2078 | } |
c0bfb650 JR |
2079 | memset(plens, 0xff, sizeof *plens); /* All prefixes, no skipping. */ |
2080 | return 0; /* Value not used in this case. */ | |
13751fd8 JR |
2081 | } |
2082 | ||
2083 | /* Returns the length of a prefix match mask for the field 'mf' in 'minimask'. | |
2084 | * Returns the u32 offset to the miniflow data in '*miniflow_index', if | |
2085 | * 'miniflow_index' is not NULL. */ | |
2086 | static unsigned int | |
2087 | minimask_get_prefix_len(const struct minimask *minimask, | |
2088 | const struct mf_field *mf) | |
2089 | { | |
c30cfa6b | 2090 | unsigned int n_bits = 0, mask_tz = 0; /* Non-zero when end of mask seen. */ |
13751fd8 JR |
2091 | uint8_t u32_ofs = mf->flow_be32ofs; |
2092 | uint8_t u32_end = u32_ofs + mf->n_bytes / 4; | |
2093 | ||
2094 | for (; u32_ofs < u32_end; ++u32_ofs) { | |
2095 | uint32_t mask; | |
2096 | mask = ntohl((OVS_FORCE ovs_be32)minimask_get(minimask, u32_ofs)); | |
2097 | ||
2098 | /* Validate mask, count the mask length. */ | |
2099 | if (mask_tz) { | |
2100 | if (mask) { | |
2101 | return 0; /* No bits allowed after mask ended. */ | |
2102 | } | |
2103 | } else { | |
2104 | if (~mask & (~mask + 1)) { | |
2105 | return 0; /* Mask not contiguous. */ | |
2106 | } | |
2107 | mask_tz = ctz32(mask); | |
c30cfa6b | 2108 | n_bits += 32 - mask_tz; |
13751fd8 JR |
2109 | } |
2110 | } | |
2111 | ||
c30cfa6b | 2112 | return n_bits; |
13751fd8 JR |
2113 | } |
2114 | ||
2115 | /* | |
2116 | * This is called only when mask prefix is known to be CIDR and non-zero. | |
2117 | * Relies on the fact that the flow and mask have the same map, and since | |
2118 | * the mask is CIDR, the storage for the flow field exists even if it | |
2119 | * happened to be zeros. | |
2120 | */ | |
2121 | static const ovs_be32 * | |
2122 | minimatch_get_prefix(const struct minimatch *match, const struct mf_field *mf) | |
2123 | { | |
27bbe15d | 2124 | return miniflow_get_be32_values(&match->flow) + |
13751fd8 JR |
2125 | count_1bits(match->flow.map & ((UINT64_C(1) << mf->flow_be32ofs) - 1)); |
2126 | } | |
2127 | ||
2128 | /* Insert rule in to the prefix tree. | |
2129 | * 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
2130 | * in 'rule'. */ | |
2131 | static void | |
2132 | trie_insert(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
2133 | { | |
69d6040e JR |
2134 | trie_insert_prefix(&trie->root, |
2135 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
2136 | } | |
2137 | ||
2138 | static void | |
f358a2cb | 2139 | trie_insert_prefix(rcu_trie_ptr *edge, const ovs_be32 *prefix, int mlen) |
69d6040e | 2140 | { |
13751fd8 | 2141 | struct trie_node *node; |
13751fd8 JR |
2142 | int ofs = 0; |
2143 | ||
2144 | /* Walk the tree. */ | |
f358a2cb | 2145 | for (; (node = ovsrcu_get_protected(struct trie_node *, edge)); |
13751fd8 JR |
2146 | edge = trie_next_edge(node, prefix, ofs)) { |
2147 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
2148 | ofs += eqbits; | |
c30cfa6b | 2149 | if (eqbits < node->n_bits) { |
13751fd8 JR |
2150 | /* Mismatch, new node needs to be inserted above. */ |
2151 | int old_branch = get_bit_at(node->prefix, eqbits); | |
f358a2cb | 2152 | struct trie_node *new_parent; |
13751fd8 | 2153 | |
f358a2cb JR |
2154 | new_parent = trie_branch_create(prefix, ofs - eqbits, eqbits, |
2155 | ofs == mlen ? 1 : 0); | |
2156 | /* Copy the node to modify it. */ | |
2157 | node = trie_node_rcu_realloc(node); | |
2158 | /* Adjust the new node for its new position in the tree. */ | |
13751fd8 | 2159 | node->prefix <<= eqbits; |
c30cfa6b | 2160 | node->n_bits -= eqbits; |
f358a2cb | 2161 | ovsrcu_set_hidden(&new_parent->edges[old_branch], node); |
13751fd8 JR |
2162 | |
2163 | /* Check if need a new branch for the new rule. */ | |
2164 | if (ofs < mlen) { | |
f358a2cb JR |
2165 | ovsrcu_set_hidden(&new_parent->edges[!old_branch], |
2166 | trie_branch_create(prefix, ofs, mlen - ofs, | |
2167 | 1)); | |
13751fd8 | 2168 | } |
f358a2cb | 2169 | ovsrcu_set(edge, new_parent); /* Publish changes. */ |
13751fd8 JR |
2170 | return; |
2171 | } | |
2172 | /* Full match so far. */ | |
2173 | ||
2174 | if (ofs == mlen) { | |
2175 | /* Full match at the current node, rule needs to be added here. */ | |
2176 | node->n_rules++; | |
2177 | return; | |
2178 | } | |
2179 | } | |
2180 | /* Must insert a new tree branch for the new rule. */ | |
f358a2cb | 2181 | ovsrcu_set(edge, trie_branch_create(prefix, ofs, mlen - ofs, 1)); |
13751fd8 JR |
2182 | } |
2183 | ||
2184 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
2185 | * in 'rule'. */ | |
2186 | static void | |
2187 | trie_remove(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
2188 | { | |
69d6040e JR |
2189 | trie_remove_prefix(&trie->root, |
2190 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
2191 | } | |
2192 | ||
2193 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
2194 | * in 'rule'. */ | |
2195 | static void | |
f358a2cb | 2196 | trie_remove_prefix(rcu_trie_ptr *root, const ovs_be32 *prefix, int mlen) |
69d6040e | 2197 | { |
13751fd8 | 2198 | struct trie_node *node; |
f358a2cb | 2199 | rcu_trie_ptr *edges[sizeof(union mf_value) * 8]; |
13751fd8 JR |
2200 | int depth = 0, ofs = 0; |
2201 | ||
2202 | /* Walk the tree. */ | |
69d6040e | 2203 | for (edges[0] = root; |
f358a2cb | 2204 | (node = ovsrcu_get_protected(struct trie_node *, edges[depth])); |
13751fd8 JR |
2205 | edges[++depth] = trie_next_edge(node, prefix, ofs)) { |
2206 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
69d6040e | 2207 | |
c30cfa6b | 2208 | if (eqbits < node->n_bits) { |
13751fd8 JR |
2209 | /* Mismatch, nothing to be removed. This should never happen, as |
2210 | * only rules in the classifier are ever removed. */ | |
2211 | break; /* Log a warning. */ | |
2212 | } | |
2213 | /* Full match so far. */ | |
2214 | ofs += eqbits; | |
2215 | ||
2216 | if (ofs == mlen) { | |
2217 | /* Full prefix match at the current node, remove rule here. */ | |
2218 | if (!node->n_rules) { | |
2219 | break; /* Log a warning. */ | |
2220 | } | |
2221 | node->n_rules--; | |
2222 | ||
2223 | /* Check if can prune the tree. */ | |
f358a2cb JR |
2224 | while (!node->n_rules) { |
2225 | struct trie_node *next, | |
2226 | *edge0 = ovsrcu_get_protected(struct trie_node *, | |
2227 | &node->edges[0]), | |
2228 | *edge1 = ovsrcu_get_protected(struct trie_node *, | |
2229 | &node->edges[1]); | |
2230 | ||
2231 | if (edge0 && edge1) { | |
2232 | break; /* A branching point, cannot prune. */ | |
2233 | } | |
2234 | ||
2235 | /* Else have at most one child node, remove this node. */ | |
2236 | next = edge0 ? edge0 : edge1; | |
13751fd8 JR |
2237 | |
2238 | if (next) { | |
c30cfa6b | 2239 | if (node->n_bits + next->n_bits > TRIE_PREFIX_BITS) { |
13751fd8 JR |
2240 | break; /* Cannot combine. */ |
2241 | } | |
f358a2cb JR |
2242 | next = trie_node_rcu_realloc(next); /* Modify. */ |
2243 | ||
13751fd8 | 2244 | /* Combine node with next. */ |
c30cfa6b JR |
2245 | next->prefix = node->prefix | next->prefix >> node->n_bits; |
2246 | next->n_bits += node->n_bits; | |
13751fd8 | 2247 | } |
13751fd8 | 2248 | /* Update the parent's edge. */ |
f358a2cb JR |
2249 | ovsrcu_set(edges[depth], next); /* Publish changes. */ |
2250 | trie_node_destroy(node); | |
2251 | ||
13751fd8 JR |
2252 | if (next || !depth) { |
2253 | /* Branch not pruned or at root, nothing more to do. */ | |
2254 | break; | |
2255 | } | |
f358a2cb JR |
2256 | node = ovsrcu_get_protected(struct trie_node *, |
2257 | edges[--depth]); | |
13751fd8 JR |
2258 | } |
2259 | return; | |
2260 | } | |
2261 | } | |
2262 | /* Cannot go deeper. This should never happen, since only rules | |
2263 | * that actually exist in the classifier are ever removed. */ | |
2264 | VLOG_WARN("Trying to remove non-existing rule from a prefix trie."); | |
2265 | } |